Hello everyone, and welcome to Deutsche Telekom's AI Investor Day. First, I want to welcome those who made their way to Bonn to share today's event with us, and even facing some unexpected difficulties along the way. I also welcome those who participate virtually and in the stream. What is our goal today? We want to show that AI has arrived at Deutsche Telekom and is already producing material benefits for the customer experience, for revenues, and for our efficiency. This afternoon, we want to go beyond the buzzwords and be as practical and tangible as possible. We expect to over-deliver on our AI-related efficiency promise for 2027 from the last Capital Markets Day, and we expect accelerating benefits in the outer years. With that, let me quickly run through today's agenda.
We start with an overview, an introductory session in which our CEO, Tim Höttges, first outlines our overall AI strategy before Birgit Bohle, our Board member for Human Resources and Legal Affairs, and Kartik Sheth, our Chief AI Officer, explain our AI operating model and how we empower our people. After this, our Board member for Germany, Rodrigo Diehl, will lead a session on AI in the networks, and this will be followed by a short lunch break. After the lunch break, our Board member for Europe, Dominique Leroy, will lead a session on AI and customer interactions, followed by our Board member for T-Systems, Ferri Abolhassan, who will lead a session on AI in the enterprise. Each of these business-related sessions will be followed by a short Q&A.
After this, the T-Mobile team, represented by Jon Freier, Jeff Simon, and Dr. John Saw, will present selected AI use cases, again, followed by a short Q&A. Finally, our Board member for Finance and currently P&T, Christian Illek, will summarize the financial impacts that we expect from AI in the medium and longer term. This will be followed by our concluding Q&A with all Board members. Please pay attention to our usual disclaimer regarding forward-looking statements. We would appreciate if today's questions are related to today's topic, artificial intelligence. Before I hand over to Tim, let me welcome our two new Board members, Jan Hofmeyr, our new Board member for Product and Technology, who will officially join us in November and, Jan, please come on stage, and Dhananjay Mirchandani, our Head of Group Controlling, who will succeed Christian Illek as Group CFO in the second quarter of 2027. Dhananjay, please.
Yeah, Jan and DJ, could you please maybe say a few words about yourselves?
Absolutely. Well, thank you very much and great to be here today. Thank you for joining us. I am very excited to join Deutsche Telekom on November 1st, running technology and product. A little bit of my background, I spent 12 years at Comcast building the X1 platform, as well as there running their networks. In the last five years, I was at Amazon AWS, first three years running all of Amazon networking, and in the last year responsible for the global telecommunication industry.
If you ask me what gets me excited, if I look across the global telecommunication, I cannot think of a better telecommunication company than Deutsche Telekom. What you will see today is they don't talk about AI as a vision, it's a reality. If I think about physical AI, I think about where the world's going, I cannot think of a better place to be. I am very excited to join the leadership team and be part of the company.
DJ? Thank you.
Yeah. Good morning, everyone, or good early afternoon. I see some familiar faces in the audience. My name's Dhananjay, formerly known as DJ within Deutsche Telekom, and I am slated to succeed Christian in May of next year as Group CFO. It's an honor, it's a privilege, and I bring five things to the table. Firstly, core finance experience in my current role and in my controlling role prior to that. Secondly, M&A experience within group development under Thorsten Langheim. Thirdly, commercial operations at Vodafone on the broadband side. Fourthly, 10 years of strategy consulting experience.
Last, and importantly, potentially for the audience, capital markets expertise as an equity research analyst. Christian has an impeccable track record, and it's my commitment to continue that, number one. And number two, among other things, as I step into this new role, I see myself as an advocate of the shareholder community and the bondholder community within the boardroom.
Excellent. Thank you, Jan. Thank you, DJ. So with that, it is my pleasure to hand over to Tim, but before you come on stage, I think we have a virtual version of you that is AI-generated to go a bit through the history of how we got here. Okay, thank you.
[Presentation]
[Presentation]
[Presentation]
Yeah. Good morning, everybody. Are you sure that I am real? Okay, at least you are.
But look, this movie was entirely made without any kind of contribution by myself. You see what is possible with AI because even the language, the pictures, the movements, everything was copying me already. To a certain extent, it is scary as well, but I can tell you the evolution of AI is unbelievable. To make that clear from the beginning, I am an AI optimist, 100% convinced. We can debate later on about all this kind of noise around what is happening in this AI world these days, but I am an optimist, and I am driving now since 2020 AI in this company with a very intensive power. We want to give you today not an overview, but of 500 projects, which we are running at that point in time. This is not the purpose of that meeting.
You should be able to answer three questions at the end of the day, and I thought about that one this morning. The first one is, can Telekom execute AI at scale? First question. The second question is, can we capture value from that? The third one is, how are we going to measure the progress and the success for our investors, for our shareholders going forward? These are the three questions we want to answer today.
We do not want to become now the AI company. I saw this headline this morning in the press and said, "What the heck?" They are putting something out there which we do not want to want. We want to enable AI to our customers and make their life so easy as possible, and we want to stay a human-led company. This is not something which we are now driving only in a technology.
We use technology to serve customers in a more simpler and in a better way, in a more efficient way. That is the purpose of our company. It should be always human-centric, both from a customer end and even from the way how we are driving it. This will not be only a machine in the future, but it will be a machine-enabled company going forward. Let me give you a brief overview about how we see the world and where we are at Deutsche Telekom these days. I think the most important thing at the end of the day is the motivation and the spirit, how we are driving things, not about all the projects. For us, AI is not just another tool.
It is a way to redesign the whole enterprise in its way of how the workflows, how the processes are getting organized in the future. It is following not only a technology, it will end in a total new governance in the way how people lead these companies going forward. To be honest, it is not the first time that we experienced that. Look, this is the story and some of you have witnessed it. Some of you were always on our side on this journey. We started in 2015 with this company, at least that is where I started, in 2014, where we said our strategy is fix the basics. To start was about enabling a company towards technology in a much better way. IP everywhere. Remember that project? We had to touch every household in our footprint with the IP migration, 100% IP accesses.
Without IP, no AI today. Service transformation, customer, and digital first. That was a statement which we made of saying, how can we help customers to get into the digital world with all the apps and all the things were rising? We drove convergence, and on top of that, we gave the first AI guidelines. I was looking it up again in 2018. In 2018, Deutsche Telekom put AI guidance out of that one. You see, we are always understanding which kind of technology developments are taking place in our universe. In 2021 Capital Markets Day, we called it the Leading Digital Telco. You might recall our presentation at that point in time. We started to dismantle our fixed-line infrastructure with the Access 4.0 layer to understand better the software implementations which we had in our fixed-line infrastructure.
We established the API layer, or we were the first one driving Open RAN of saying we need a new technology to control the value chain of our things.
We started starting digital experts, scaling digital experts for our ecosystem. By the way, almost 30% of all employees are digital experts within Deutsche Telekom. We started with the first digital services, which we rolled out across the whole universe of Deutsche Telekom. 2024, we called it data-informed and AI-enabled digital-first company. To be honest, this was a little bit too academic. But it was heading the way. We were techno enthusiasts at that time. Mike Sievert and the U.S. team started these initiatives, but we started with all, let's say, the developments on AI in 2022, where OpenAI moment took place. Since then, we are on a journey. Having all the investor calls with you guys, I know all your questions. You want to understand what's behind that. We have 500 projects. We have started the sovereign AI cloud. We have new revenue potentials.
We have built an AI factory inside, AI competence center, as we call, with supporting the organization where the competencies are not given. We have started a rollout program across the whole organization with trainings for almost every employee. Birgit will go into this one later on. Now it's the next time leading AI telco. I even think best network will be best AI network because it will enable. Without our networks, no benefit of AI at all. By the way, the kill switch for every AI sits in the network. If we're switching off the network on mobile and fixed line, there's nothing happening. No workflow is going. Being it on the sensor, being it the robotics, being it, let's say, on the cloud access, the network stays a core competence of every AI functionality which we're going to have.
We are part of that journey with our infrastructure, this is what we are talking about today, how we can drive this next level with our telecom infrastructure going forward, autonomous AI-driven services, and connectivity for all the agents which are in the infrastructure. That said, we are thinking there are three big dimensions which we have to look at. The first one is how is the network changing? The second is the customer. How is the customer impacted? How can we make customers' life better? The third one is how we can monetize AI functionalities within the enterprise era. Interesting what you observe immediately, the whole company is affected by AI services. On the network side, by the way, AI is taking place in access, edge, and the core.
We will have something in this infrastructure which we can monetize on top of what we have today. I would not call it immediate tokenization because nobody knows what a token really is. We have to monetize the traffic which is generated with AI in this one. We can, at the same time, reduce the run cost for our infrastructure with autonomous agents. We can improve the resilience of our infrastructure, we have new revenue streams which we can build into the infrastructure. On the customer side, it's very easy. The cost to serve by bots, the cost to serve by agents, the cost to serve will go down. We will support our human people with the tools on the AI side.
By the way, at the beginning, we thought we can only keep the same NPS with the agents that we have with the human interaction. What we found out in projects, we will talk about that one today, is that we can even improve the NPS. Give you an example. You have never heard about waiting times at Deutsche Telekom anymore. It's over. There are no waiting times because we always have somebody who is interacting with our clients. We are able to increase the interaction with our customers with last best offers, that is driving the retention to better scales. The last one is the enterprise area. The enterprise area is we have a distribution which is unique with T-Systems, with our B2B arm. EUR 15 billion B2B revenues already today. We have contact to all business customers in Germany.
Give you an example, 60% market share on the mobile connectivity and the fixed line connectivity. We can upgrade these customers. I'm doing three to four customer interactions every week, I can tell you we are only talking about this, nothing else. The connectivity is given for them. The only thing what interests them, how can we help them to get into the AI economy? That's a big opportunity for us, which we have to leverage going forward. These are the three dimensions we're going to discuss today in all depth. How are we going to drive the KPIs behind that? There is an underlying architecture, which is the enabler, the enabler is super important to make all these kind of developments successful.
I'm not going into this one in detail, but how can we organize an organization where you don't have an assistant, which is AI, where you don't drive one agent, but where our people are running multi agents in a kind of safe environment? This is the big challenge which we have. This is the team agent logic which we are driving and which is already taking place in a lot of areas. But how can we make that not only possible but even secure going forward? We have a governance established around our organization, which Kartik is going to explain in a second. Let me move on to the revenue potentials which we see. By the way, on consumer AI, we will see the next iPhone moment coming. I say iPhone moment, I do not know whether it's going to be Apple.
But now having seen 18 versions of the same smartphone, we all know in this room there is a moment coming, I do not know when it's coming, where this is over. Because you do not need this kind of device where you have to tap apps, where you have to go through a complex administration, where you have to write your intent. The future is going to be voice native and there will be a moment where a new device is coming, where you have an iPhone moment and said, "I do not need this smartphone anymore." The question is, when is that going to take place?
We have our theory on that one, but this is a big opportunity because here, for the new devices, we're going to need the same safe, smart, secure support for this infrastructure, and we have another opportunity as well to build a new competition, a new ecosystem in this highly subsidized hardware world in which we are suffering today. Second, we are going to go into infrastructure. By the way, I was in the Valley recently, and I can tell you to each of the guys being, let's say, Dario, being it Sam, being it, let's say, Sundar, always had the same observation. Welcome to the telecom world. Because we're sitting on EUR 300 billion of capital employed, and on top of that, we have to have long amortization rates. These are guys going in the same industry like ours.
It's an infrastructure-heavy industry where they have to amortize over economies of scale the big business case they're doing in data center, doing in their large language models. They need scale. They need distribution. Otherwise, the whole investments into these infrastructures, which are unbelievably high. Yesterday, I read an article that it's now the Anthropic IPO is creating 9% of the GDP of the U.S., otherwise the business case of $2 trillion would not work. This shows you something that the utilization of these capital-intensive industries, like telcos, is now being copied in this world. This logic is something we deeply understand. Deeply understand. Therefore, we understand if we are winning. This is, by the way, the rendering of the gigafactory we're working on, that Ferri is going to talk about, because this is something we understand.
We're going to build sovereign infrastructures and data center infrastructures beyond, let's say, what we have done so far. We see an opportunity here as well. The third one I mentioned already is the question, if you have AI everywhere, if the AI GPUs or inference is sitting not only in big clouds, but sitting in the infrastructure which everybody is thinking about, you have AI at the RAN. You can optimize compute, and you can optimize the connectivity latency. Jitter is becoming super relevant for robots and other things. You need AI capabilities at the core. This is something which we are able to drive, and we will have inference data centers in the telecom infrastructure going forward, which is an additional opportunity for us to monetize. EUR 1 billion extra revenue is something which we have in mind. To be honest, this is a conservative number.
I can tell you one thing, I do not want to commit a bigger number because you will anyhow not believe it. Therefore, we go with EUR 1 billion, but our internal plannings are much more optimistic than what we have laid out. Let's start with a billion and then we will see, but the opportunities along that value chain are unique for a telco. Now you ask, by the way, why is Deutsche Telekom winning in this? Do they have a right to play in this world of Anthropic and OpenAI and the Geminis of this world? Do they have really a right to play? They don't have a frontier model. They don't have the data center at scale. They don't have the economies of scale like the Americans have today. They don't have the forward deploy experts.
They don't have, let's say, the agent models as these guys are offering, so they get cannibalized. To be honest, think this is not our game. We don't want to compete in these games. These are our partners. By the way, they're going to be our partners because they need, for their economies of scale and their heavy-loaded infrastructure, they need partners like us. Our play is totally different. Our place is where trust matters. That is, let's say, where sovereignty matters. That is the area where secure and access matters. This is where a telco like Deutsche Telekom has a right to play. By the way, look at this. Do we have a right to play in this area with our customers today?
If you ask today our customers, "Are you going to go to MUSE or Instinct?" I can tell you the first thing what everybody says to you, and I discussed it with my kids on the weekend, by the way, 50 kids because there was a birthday party. We discussed it up and down. They said, "By the way, can I trust my data? Where is it going?" Who is the provider of that? What is happening? Is that another advertising model? I can tell you, trust matters. We are number one telco brand in the world, by the way, by far leading in our European markets with regard to trustworthiness and the competition. We are reliable partner of governmental services.
I can give you a long list from Bundeswehr to the Bundes app, which we are building, where the government say, "We want to work with you, with Europeans, with Germans," rather going to the big American companies. Context and scale. I was very surprised how interested the big AI companies are to get the distribution of Europe. They all want to partner with us because they want to get access to our 200 million customers, by the way, contract customers. They want to get access to the distribution of our B2B network. On top of that, we have 100 billion calls on our network, 100 million contract center interactions per year. Think about the intimacy and the understanding about each of the customer, which we have. AI is enabling us to understand that much faster, much better.
Unmatched customer access and insights on both sides of the Atlantic. Infra asset. I do not want to go into this one. You all know that we are perceived as the one who is able, with this balance sheet, with his capabilities, with his systems in the back, to drive data center infrastructure going forward. We are already doing it quite successfully. We have the ecosystem access. We know everybody in this world. We are the only telco enabled in America, which is very much driving this AI world. We have on top of that an arm, which is DTCP. We are invested in Net and BF. We are in Lovable. We are in Black Forest Labs, in NScale, in Decagon.
These are the ones which you need to enable all the AI services, and with our DTCP arm, we are invested in these companies. On top of that, with this, we have access to their technologies and using them. So we have a unique kind of ecosystem which we have generated. I have no problem with any of them to be deprioritized in each of these companies. We have a preferred access. The last one is people. 250,000 AI enablement since 2010 with these. People asking for education, people asking for trainings. Interesting-wise, the mid-management is super enthusiastic about AI. The most important thing which you should keep in mind is we have in this company not a panic about AI, not this European sentiment that AI is a bad thing.
We have created a movement here, and I am so convinced that this is the biggest advantage which we have compared to other telcos. Because we do not have these discussions about, "Oh, this is all a bad thing." Yes, we have them from here to there. But there is something that people are pulling AI everywhere. It is more the question of organizing this request than rather making them hungry to AI. This is, by the way, the sustainability argument of this one. Look, we are going to drive this with renewable energy. This is what we are doing. By the way, even if we do that here in Germany, when we talk about the gigafactory, we are focusing on our PPAs and the way what we can using.
It's an additional energy consumption, but we want to try that we get into net zero emissions, even for the increased amount of energy which we're going to use for AI applications. We are following our principles as well on here, knowing that this is becoming an additional challenge for us going forward. Our biggest challenge is, by the way, not scope 1 or 2, it's scope 3. Second, trust and security. I cannot go into this one, but we have to deep dive on cybersecurity at a scale. We saw 20,000 DDoS attacks last month on our German network. By the way, it's like tanks running over streets. Oh, where are they going? We see them on our networks. We can, if the legal requirements are given, we can stop them, at least the majority of them.
Our network sees the attacks, five billion security-relevant data points, which we're analyzing. By the way, it's only possible with AI to get control of that one. And we have already a 99% accuracy AI detection of false positives because there's a lot of, let's say, irrelevant stuff in it. And a resilient infrastructure for crisis management. By the way, we had the head of staff of the NATO. We had two generals in our board room. KRITIS, DORA, all of this. The government is deeply relying on the infrastructure in this kind of difficult geopolitical times, and we can talk about what we're doing about that one. I think our network resilience is super relevant for Europe and especially for the German government. Today, it's not a vision anymore. I do not go this. This is what we do. 95% RAN guardians, faster incident handling.
42% AI-generated code already in that company. 2.6 million deflected calls by the bots which we are using. We have 10,000 GPUs running in our data center in the Industrial AI Cloud in Munich. Or let's say all the use cases which we have for B2B customers. These are not inside cases. These are 400 cases which we are doing and driving with our B2B community today. Second, AI innovations. Building the network of 6G. Who can build the most resilient 6G network in the Western world? I leave the answer to you because if you look to U.S. and Europe, there's only one player who is present in both of the markets. And on top of that, we have in the U.S. quite a lot of successes already, 55% call reduction. The Net Promoter Score is up by 12% on expert assist.
And we have as well 30,000 antennas who immediately adjusted themselves automatically during winter storms. When there is a fall off of some antennas, the propagation, the antenna, the beam-forming of the antenna is changing automatically to cover more areas, splitting the spectrum towards more clients. This is what we achieved, let's say, automatically in the U.S. But that's not enough. We have audacious targets going forward. 100% of calls initially should be handled by AI. This is 2030. 50%-55% reduced calls and chats versus 2023. 70% retention and tariff changes via digital services, and an improvement of another 5% overall in the transactional Net Promoter Score. Second, on the network. 10%-15% network gross savings. Autonomous networks, we call them level four, and an improvement on the CXI, and we talk about that later on, to four. IT and software.
Up to 60% of our software generated automatically with AI. Software developer productivity improved and 70%-80% code written by AI, and 15%-20% gross savings across the whole organization. On financial impact, we have said that we are going to have EUR 800 million. We can confirm today that in 2027, we will have EUR 1.1 billion gross savings by that one. So we exceed the number by EUR 300 million already today through the implementation of the project which we are having. We believe on top of that, by 2030, we are going to have EUR 2.5 billion gross savings from AI and automation within the footprint of our European organization. Last but not least, there is a EUR 1 billion additional revenue potential. I mentioned that already. So these are our audacious targets going forward. By the way, that doesn't mean that they are visionary.
They are just implemented in our planning process, on which we are working out right now. Brings me to the end about the story. We laid out in the last Capital Markets Day 2024 that we are going to accelerate our flywheel by data and AI and by global scale. By the way, this is what we are talking about today, the acceleration about data and AI, how to accelerate a physical AI investment infrastructure, with significantly lower latency, supported by own sovereign clouds. More experienced and individualized products, segmentation along the customer journey. Leveraging this in a global scale, starting at least here on the European side, but improving in our collaboration with the U.S. Bringing more people behind that on an AI side, you know that everybody is enabled. 120,000 license of OpenAI are implemented or getting implemented soon in the organization.
Creating higher efficiency across every single workflow of this organization. To be very clear, we are just at the beginning of redesigning the workflows and the governance of this organization. With this, improving our margins and the revenue pool of these organizations. While telco is getting more and more saturated, we need new revenue streams. With, let's say, the infrastructure, with the B2B opportunities, we see revenue potentials on service revenues going forward. That's the way how we are going for it. We are, again, AI optimists at that company. With that, we today want to show you not just PowerPoint slides, but for each of these elements of the three areas, we want to deep dive and to understand how we are going to do this.
Before we go into the showcases or into the projects here, I want to ask Birgit and Kartik to give us an understanding, how are we doing thing? This is about what we want to do. But the big question is, having an organization that scale, how can you implement that? That it is not ending up in a total chaos, that you have a motivated workforce behind that. With this, I hand it over to Birgit and Kartik. Thank you.
Thank you, Tim, for setting the stage. Actually, two years ago, I was talking here at CMD about three things that set us apart: the best team, a unique culture, and a transformation DNA. All three are still valid today, and I believe they are more important than ever to now enter the next phase of redesigning the enterprise. As Tim said, I am going to share the stage with Kartik, our Chief AI Officer, and the two of us, we will talk about 20 minutes about the how. How we are shaping our AI operating model, how we are reimagining work, and how we are moving from AI adoption to AI impact. Let me start with our people and that movement that we created. Yes, we have built a real positive AI momentum at DT. That always starts with leadership commitment.
It was actually Tim who made AI a top strategic priority and also a bit of his personal learning priority. That sends a very clear message, leaders must own the AI transformation. This spring, we spent three full days with our leaders focusing entirely on AI. We got inspiration from top AI companies, but we literally went back to AI school, developing our own agents and small applications. Leadership commitment is essential, but of course, transformation only happens if we see it in every team. I can tell you, whenever I am with operations teams these days, I can tell you it is real. Recently, I met this builder's service team in Dortmund. They had just redesigned the whole fiber ordering process for property developers from scratch using AI. One colleague had actually built an AI-run wiki, which is now being used by the whole team.
That is actually the spirit I experience everywhere. Today, we have thousands of colleagues active in AI communities. They are sharing ideas, they are testing solutions, they are learning from one each other. This is Exploration Day, a grassroots initiative, and Kartik actually ran into this on his very first or second day, and he told me, "I was so impressed with the quality of that session, and especially with the fact that the employees had actually self-organized that event." The numbers show the momentum. 83% are using AI regularly now for their work. That is what the employees tell us. Our ambition is very clear, 100% AI for everyone. We provide the right tools. two years ago, we demoed AskT. We have now hit the 10 million conversations across the company just last Friday. We are rolling out ChatGPT Enterprise. We are rolling out Microsoft Copilot.
Our developers use, of course, specialized tools, our legal team's Harvey, and our leaders actually now have an AI coach called Nadia. But adoption, of course, is not the finish line, it is just the starting point. Since 2023, we had more than 250,000 enablements. Tim just mentioned it. More importantly, our colleagues actually report substantial productivity improvements and use that time to actually improve the quality of work. That is what we are ultimately looking for. Better work, better outcomes for our customers, not simply more usage. With that, I am handing over to Kartik, who tells us now how we turn that momentum into redesigning the enterprise.
Thank you. Thank you, Birgit. She spoke about how we have AI momentum across the company and the positivity within the organization. I will talk about how we convert that momentum into measurable impact. This is not like a regular transformation, where you have a three-year plan and you execute to that plan. This is a world where the ecosystem changes, the technology changes, and everything changes very fast. How do we navigate that change and still deliver impact continuously is quite the challenge. I will walk you through instead five postures, not a strategic plan, but specific postures. It helps us define how we work in a world where what we do is changing and will continue to change. Let me walk you through five of these. The first one is our data.
As you can imagine, AI is only good as the data that is available to it. What we have done through a unified data platform is find a bridge between the old world and the new world. So it has a reference data platform, which is AI native, built from scratch, suited to AI workloads. Over that, we have two other layers, a data marketplace and a semantic layer. The data marketplace helps an agent identify where in our current federated system of data something resides, and the semantic layer helps translate that to business logic and vice versa.
Today we have a series of different data platforms that exists across the organization built for different segments, different purposes, and the AI agent is able to use the UDP to navigate that seamlessly and all the migrations and all the improvements, transitions that happen are completely transparent to the user. They do not have to worry about it. Over a period of time, this becomes the brain of Deutsche Telekom. This becomes the differentiator because the better the data, the better the information in there, the better the AI agents have in terms of their runtime. The second Posture that I will talk about are our partnerships. Tim spoke about this. This is a thriving ecosystem. We have companies here which are worth not just billions, but even trillions of dollars. They are together pushing the boundaries of technology at a speed we have never seen before.
With each of the logos here, we have multi-dimensional partnerships. That means in some of these, we have investments, in some of these, we go to market together, in some of these, we have infrastructure that we work on together, and this gives us strategic depth. I will give you a couple of examples. With Lovable, we are the first European client, and this is based on feedback we gave them, where you can build something on Lovable but deploy it internally on our premises where we can connect secure data to it. Rapidly accelerates development. With Decagon, we not only get technology, but also practices because they serve customer service clients across the world. Given the feedback we gave them, we have created a tool for mining those insights, which helps us improve our customer service practices, not just the technology.
With ElevenLabs, you saw that already at MWC, that we are working with them on integrating voice models, translation services into the network. The third posture is a defined AI operating model. Tim spoke about how we want humans to lead the teams of agents. AI is in everything that we do. The lane 2 is pretty much about that. It is about us providing the infrastructure, providing the FDs and resources so that every business leader, every functional leader can automate, can build agentic automation bottom up. Examples of this would be a financial forecast system in Hungary or a tool to automate RFP actions in Greece. This is helping enable AI in everything we do. Lane 1 is for our big bets, and for a telco, GenAI is probably the closest thing you can get to a silver bullet.
It allows us to optimize our CapEx, reduce our OpEx, grow our revenue, all at the same time while improving customer service. A small number of those bets have to be driven down across the group, and I will talk a little bit more about that in the next couple of charts. I am going to talk about this slide backwards. This is really how we implement the small number of big bets. First, we have selected these, in sales and services example I will walk you through. We then redesign to make sure that we are building for the future and not for the past, and I will explain that with an example. Then we build, deploy, and scale exactly in the way an AI-native company would do. A small group of AI-native builders using an AI SDLC builds the product.
A group of FDs goes into every country, every segment to help implement that and leaves behind the capability, so that future agents development does not need to be a tech change. I will talk a little bit more about redesign with an example. So in sales and service, we have a tiger team that has representation from six different business segments, as well as products and technology. They look at every single thing that we are automating and saying, "Should we automate what we have or should we try and redesign, use that opportunity," as Tim said, "to redesign the enterprise?" I will give an example of authentication. There are many others that this tiger team looks at, some small, some large, but I will use a small example. Authentication, which is what a customer needs to identify themselves on the phone call.
Over a period of years and across nine countries, we had developed 14 different methods of authentication. The tiger team looked at those de novo and said, "What do we need going forward?" They came back and said, "Look, we only need three of these. These are future-forward technology-wise. They are customer friendly. Redesign the processes so that those three are very well built." From a technology perspective, that allows us to build fast and scale up fast while giving a better customer service. Now, imagine this happening hundreds of times in sales and service and then across other functions as well. I come back to what does that mean in terms of the stack? A flexible AI stack is our fourth posture. Our stack needs to adapt to different technologies, different concerns that are coming in, partner changes, and also changing cost structures.
At the top layer, you have agents which are built on AI products. Many of these products, you will see these in action today. These are products that are live today and in use. Below that, we have the data model that I already spoke about. Then we have switchable harnesses. I will spend a second on this. Switchable harnesses allow us to change not just the model, so the same application, but depending on what is working well, new technology coming in, voice quality allows us to switch the model. Not just the model, it also allows us to switch the partners. For example, in sales and service, depending on whose technology is doing better and in what use case, we can dynamically route a call to one of multiple partners. It is harness switching, not just at the model level, but also at the operator level.
Guardrails are becoming more and more critical. Then you have a selection of models and GPU compute that helps for every use case. What that does is it gives us maximum reuse of components. We are not building it all over again. It gives us minimum lock-in. We can switch when we need to, and we have done a couple of those already. A strong guardrail set because that is going to be the foundation of the trust that Tim spoke about. Then we constantly optimize for ROI and costs because these things keep changing. That is what the flexible AI stack looks like, and that brings us back to the overall grid that I spoke about. Probably the most important one here is the human in the lead. Besides trust with customers, there is also how do we build this capability with our teams?
If data is a differentiator, this will be why we win, because our team is different. Our setup is different. Our network with partners is different. With that, I give it back to Birgit to take it forward.
Thank you, Kartik. That was a glance into the machine room of our AI transformation. Ultimately, redesigning the enterprise is about reimagining work. From the people side, we are supporting that with three things: redesigning the workforce, supporting with stronger capabilities, building up different roles, and of course, AI for everyone, and also in our area, powered by the best tools. I get asked a lot, "What will AI mean for our workforce over the next 10 years?" The honest answer is, I do not know, and no one does. What we can do is we prepare systematically. What we are doing right now is, together with the business, actually breaking down roles into tasks and tasks into activities. Then we are asking ourselves three questions. That task, will that be human-led?
Will it be made faster and better by our people supported by AI, or can it be automated? Take a software engineer to be concrete. AI can already write code and run tests. Work is redistributed between human and AI. The future engineers, they define what to build. They direct, orchestrate agent teams. They review, judge output. They ultimately own the outcome. The role doesn't disappear, but it's moving upstream. Running this exercise across the company, we clearly see where we need fewer resources, different skills, and entirely new roles and capabilities. Yes, of course, AI will increase our productivity, and we will accelerate our ambitious workforce transformation, always doing it in a socially responsible way. The more important insight, though, is how skills and roles change and where we will need to hire and build new capabilities. Kartik just described the AI operating model.
We translated the two lanes of impact delivery jointly into the capabilities and roles that we need as its backbone. For the big bets, we put on the biggest and the best expertise that we have with the Tiger teams who identify the biggest value of pockets, the AI-native product teams, and then the forward-deployed engineers who actually deploy, configure, and help implementing it locally. We need strong AI capability inside every team, like in the Dortmund example, with AI leads that drive change in their domain, and we have AI builders who are able to develop smaller AI-powered solutions themselves. Taking my HR team, I now have a handful of leaders with a very strong technical and product development background and have a growing number of, we call them techies, that are building AI products such as, for example, our new onboarding app. Then AI for everyone.
It's very clear this is no longer simply about giving access to everyone. Everyone needs to have the confidence to actually use AI, know when to use it, also when not to trust it. Let me pause there for a second. Our skilling goes far beyond pure technical AI fluency. Critical thinking, judgment is one of the most important skills of the AI age, and that's what we teach as well. How do we put that into practice? Our ambition is simple, being the best and preparing our people for the future. We're entering the next phase. We broaden access to external certificates. We already have Create-a-thons, AI wizards who support the teams to actually redesign the work and solve real business challenges. We're investing in our leaders.
Starting 2027, we have a first cohort of leaders actually joining T-Mobile US on an AI academy, and Nadia is being rolled out. We know that we ask a lot of our people these days. We ask them to continuously embrace change and invest into their own skills. There comes a responsibility with that, so our North Star is employability, helping our people thrive in their role, prepare them for the future, wherever that may be, inside or maybe even outside Deutsche Telekom. Our most important tool to support that is actually Growth Hub, already live for almost 120,000 colleagues. Let's have a look into that. Let me close where I started, with the best team, the transformation DNA, and our unique culture. Today, we call that culture T-Style. We're convinced that in the age of AI, culture matters even more.
We can redesign roles, build new capabilities, we can provide the best AI tools, but none of it will scale without the right culture. Culture is actually not a soft factor, it's an execution factor. Transformation only accelerates when we have leaders and people with a leading attitude who are questioning established ways of working, experimenting with AI, and proudly share what works and collaborate. That's what we will strengthen with T-Style: performance, the un-corporate, and collaboration. AI will fundamentally change our enterprise. Some things remain the same, our compass, our values, our strong purpose, and our ambition to be the leading digital telco. Here is our formula for being leading in the AI age. The best team meets the best technology powered by the best culture, our T-Style. Thank you.
The staff is here. Okay. Thank you. Thank you, Birgit, thank you, Kartik, and thank you, Tim, of course, for setting the stage and showing how we want to make the most of the AI opportunity at Deutsche Telekom. We don't have a Q&A now, but Tim and Birgit will participate in our final Q&A at the end of the afternoon. While I speak, we are preparing the next session. We talked about tangible impact. We produced something very tangible here, as you can see. The next session will be AI in networks. Rodrigo Diehl, our head of Germany, will lead through the session, starting with our strategic thinking for the network of the future before introducing three demo cases that showcase the contribution of AI towards this vision. Rodrigo?
Hello, everybody. Good afternoon already. Good to see some familiar faces. I will talk about, as Hannes said, AI in networks, but I will do it together with some colleagues. What we will show you is our vision of how the network of the future will look like, but also how AI is changing the way we operate these networks. As I said, I will do it with some colleagues. We have here Alex Jenbar, who is our CTO for the German operation. He's part of my management team. We have Armin Sumesgutner, our Chief Network Officer in Europe, and we have Arash Ashouriha, who is our group technology officer. I've been around for more than 20 years in this industry, but I believe that today some things are changing, and I believe that our networks are fundamentally changing.
There is four reasons why I believe that this time it will be different. First of all, server farms across the ocean will not be fast enough for a lot of AI use cases and workloads that are emerging. That means some of that intelligence will need to move closer to the device, closer to the end user, closer to the edge. Second, we live in unstable, unpredictable times. I also believe that the world of a few large clouds in few places of the world, that that logic will not be enough, that ensuring trust and sovereignty locally will be relevant. Our networks, the infrastructure we build and we run is local by definition. Third, 6G. We have seen the different Gs, but 6G is not going to be just another G. 6G is not going to be only about transporting data in the network.
6G is also going to be about sensing that information, interpreting that information, acting upon that information, and that is a fundamental shift versus what we have seen in the last couple of years, where the network was mainly a transport layer. Finally, network orchestration. We are seeing a multitude of access technologies going forward. What's going to matter is how those different access technologies are orchestrated, are managed in a seamless way. These trends will also enable new services and new chances for us to monetize these services. Number one, Tim mentioned the post-smartphone world. We will have a new iPhone moment. Is that new iPhone moment not already happened? How will that world look like? AI agents that have memory, that understand context, that act upon. AI agents where the interface will probably be the voice.
I think the days of using our fingers to talk to computers is over. Voice will become the new interface, and this will bring new hardware form factors, earphones, glasses, pods that we will carry. We are really, really good at scaling, distributing new technologies, new forms of hardware, especially when they come with connectivity, and these devices will all come with connectivity. Second, physical AI. For a lot of the AI use cases that are coming to run in the real world, and with the real world, I mean outside of factories or outside of campus networks, the networks of today will not be enough in terms of jitter, in terms of latency, in terms of quality of service.
By the way, we were recently Silicon Valley talking to some entrepreneurs and they were telling us, "Today, the biggest limitation I have is that the latency or the jitter or the quality of service is still not there, what I need for my robots or my physical AI to function in the real world." That will be an opportunity for us, and that will be an opportunity to further differentiate through network quality. Third, network of networks. A network that is seamless, that is always on with a level of security, with a level of trust. Four, intent-based services.
The network will have intelligence, and that will allow the network to see intent, to understand what physical AI, what different devices demand and require, and will be able to adapt the service, the quality of that service, the slices, the speed, the jitter, the latency to those specific services. Finally, sensing. Think of a world where a network will be able to see and view as different devices attached to our network, where our network will allow us to build a digital twin of our real world and the use cases we will build upon it. This is how the world looks today. If you think about AI coming, if you think about super intelligence coming, most of the brain power is in data centers, in AI factories, and some of the brain power is in the devices. The network is mainly a transport layer.
That will change. Because if this system has to evolve, and if this system has to deliver to the expectations, it also needs to evolve not only its brain power, which we are seeing the trillions of investments that are going into the brain power, but it will also need to evolve its sensory system, its nervous system. That is what the network will be. As we build intelligence and AI into the network, the network itself will become intelligent. A system can only really function if it has a proper nervous and sensory system. This is, I guess, why you see companies like Nvidia investing into companies like Nokia, because that intelligence will be built into the networks.
John Saw will dig deeper later into some of the topics that I touched, but it is not only about the network of the future, it is about how we will operate the network of the future. Our vision is of a fully autonomous network operation. Some of the examples we will show you today hopefully will show you that this vision is not as far away as we think. Why we do this, we do this to differentiate. As I said, I believe that these capabilities will break the trend. Networks will not become further commoditized. Networks will differentiate more on quality and on some of these capabilities that we will build. Second, of course, this is our biggest cost bucket, both on the OpEx and on the CapEx side, and autonomy AI will allow us to extract efficiencies from that.
We build everything on data, and we build everything on customer data. What we have implemented this year across all of our operations in Europe is the first foundation layer for that. We call it our Customer Experience Index, CXI. We measure today for each and every customer attached to our network, the real-time network experience across a multitude of indicators. We have correlated that with NPS. We have correlated that with churn. We know that these are the things that matter to our customers. Based on that, we make our investments. Based on that, we decide where to put capacity, where to invest in the network. We can measure after a change if the experience of that customer has improved. We can actually measure and define if a customer has an opportunity for an upsell or if we need to implement a churn retention measure.
This also brings monetization. This is not theory, this is real. This year, a couple of weeks ago, in Germany, we won the prize for the most consistent mobile network in Europe. This is a prize that we won among 120 mobile networks in Europe for the first time. We had never won it before, and we did it building on this foundation. This is how we optimize now, every day, our network. I believe for all the reasons that I mentioned before, that consistency, assurance of quality, will be key to enable the physical AI world outside there. One element that is, of course, particularly important to me, as this is our biggest project, I call it our generational project in Germany, is how this also translates into efficiencies for a fiber network, for a fiber build.
What you see here is our investments into fiber are rather constant over the years. What we get out from that investment is increasing, and these are the efficiencies that we get out of the fiber rollout and that we reinvest into monetizing our infrastructure. As an example, today, with a similar CapEx envelope, we are connecting more than three times as many customers as we did only three years ago. This, as you know, where the homes connect and the homes activated are actually the expensive part of our network build. We are using these efficiencies to reinvest in the network and gain further momentum in our fiber rollout along the lines of our flywheel that you know.
Now, to show you that this is not just PowerPoint, to show you that this is very concrete things that today are making a difference for Deutsche Telekom, my colleagues will guide you through three cases. First, Alex is going to show you how we use AI and automation for our fiber rollout in Germany. Second, Armin is going to give you an example of an AI capability that is today live in our network and that generates already today many efficiencies, but even more improvements in customer experience. Finally, Arash, with MINDR, is going to give you a glimpse into what is coming. MINDR is an MVP that is about to go live, and this is where I believe soon network operations will be. With that, Alex, I welcome you here. Thank you.
Thanks. Good afternoon. Rodrigo showed us the future of the network. We believe in fiber. Fiber is the foundation of our belief in the future network. It is very important for us. Building fiber is not that easy. It is a big process, a very complicated process. We started with acquiring permission, selecting a vendor, selecting an area, going through all of the work, having construction teams outside. It is a very complicated process, which requires a lot of attention. In order to be on time, to have the right quality, to have also the costs under control, we implemented AI all over the value chain of our AI. From the area selection, to planning, to building, and documenting what we are going to do. It ensures us a smooth operations in our fiber journey.
Let me give you a sense of the scale we are talking about. Today, we have over 14 million homes passed already in Germany. 2,000 construction crews are working outside and 4,300 areas, and we are building 16,000 kilometers fiber every year. This gives us a challenge. This is a massive operations, and this gives us a challenge. The challenge is how to maintain control over this massive operation across the country, a large country like Germany, with this scale we are talking about. We were thinking how to find out what is going on in the field. Of course our teams are going out. Our teams do have the quality controls. They are going out and looking at the construction sites. However, we cannot be everywhere.
It's almost impossible to be on every site with all our crews and to maintain the control of what's going on because we also work with a lot of partners across Germany. We asked ourselves how to address that issue. AI is the answer to that issue. We ask ourself how we can utilize AI to bring the construction sites in a digital way, controlled way, analyzed way, into our offices, into Bonn or our offices across Germany, where we are going to build fiber. That's where we talk about the digital construction management. We created a digital view of what's happening in the field outside. AI helps us to process, analyze, compare the information at scale, and check the quality. Monitoring the progress, automating documentation, and very important one, to protect the investment for accurate billing.
Because we are getting charged from our partners, we need to protect the investment we are putting into the fiber. It's helping us very much in that area. Last but not least, if we use that, to identify the problem before it occurs. Because once you dig something into the ground and you close it, and you have not identified the problem, it could become very costly. That's why we have the digital construction management to see what's going on, to have a deeper view into our work across Germany, and to identify if there is an issue to not having expensive surprise at the end. Before talking too much here on PowerPoint, let me show it to you. We are together with our head of construction management here. Okay, please join me on stage. We will show you a demo here.
We have built here for you, this is a copy of our trench outside. You have a trench, the homes passed, and we have also the buildings prepared here for you. You will see live how it will happen. We have a device. All of our crews outside, regardless if they are internal or external, they have this device connected using a GPS signal and videoing and photographing all the construction sites. Now we are videoing the trench and also the buildings prepared part and transferring this directly into our platform. The platform we are using is a platform which is open to all the devices outside on the market. Regardless if the device has been provided from us or from our partners or whomever, we can process that one into our platform. This is very unique on the market.
No one else has that, and we are the only one using GenAI in the background to analyze what's going on in the network and outside in construction areas. We are uploading here the data real-time and going. We assume now the data is on our platform. Just please can you take us through and then I will take you through what happened and what we have seen. Please.
Yes. Here you can see our digital construction management platform. You here see the overview of the AI quality checks, which we just run with a device here. Here we have the results, and we see three of those five AI quality checks are positive and two are negative. For example, we check if the buildings prepared is properly done, so that ensures us for a fast customer activation. This is green here in this case.
Let's not focus on the good ones, because as Alex just explained, it's very expensive if we do not find the errors in time and cannot resolve them before the trench is closed. Let's focus on the depth measurement. I'm going here into the scan, and you see just a 3D model of our trench, which we just scanned. You can see here in red is marked that the depth is not accurate. You see the depth is too shallow here.
Exactly. That's where the problem starts. Imagine you are outside and the company who's working for us is digging the cable in. If the depth is not in the quality it should be, in the right size it should be, according to the permission, and once it is closed, it could create a lot of issues afterwards. That's where we try to protect our investment and how we are working on the field outside. It gives us immediately a snapshot what's going on, so we can address our partners and say, "Listen, this is what you have to change." That's which gives us an advantage compared to all other alternates on the market. Thanks. Go to the next one, please.
Yes. Let's also look at the Kugelmarker. Kugelmarkers are also very important, especially to find our infrastructure in case of an error afterwards. Here you can see the AI detected that there is a missing Kugelmarker here at the buildings prepared. It's lying here at the edge, but it should be here where the connection pipe to the buildings prepared is in the trench.
To explain that one, of course, the homes passed is going towards the street and every building is prepared, connected, and the Kugelmarker is exactly there where we go to the building. If it is not there and one day we would have an issue with the cable or we need to open it, then we would open the wrong side or the wrong place on the street. It creates more cost, permissions, time, and a lot of impact to the customers. That is why it is also very important to identify these kind of issues beforehand, before we are closing the ground and going to connect the customers. Thank you very much.
Thank you.
Let me give you also numbers. We have with this 100% transparency, 15% quality improvement, 20% corrected invoices. That is a lot of money because we are spending a lot of money in our fiber operations. So 20% of corrected invoices protect us, protect the investment, and 100% automated documentation. The numbers speak for themselves and the impact speaks for itself. Thank you very much.
Good. Thank you, Anke, and thank you, Alex, for this exciting demo of how we use AI to increase our fiber build efficiency. As Alex has said, we spend so much on fiber that every percent of efficiency is a big number. You understand numbers, so you can do the math. Next up is Armin Sumesgutner. He will talk about how we use AI in the mobile network to proactively avoid congestion and improve the customer experience. Armin? Great.
So we move right now from the fixed network to the mobile network, and we move into something really exciting. Of course, in modern network planning, we typically plan for customer experience. We try to get the best customer experience at a certain spot for the people. That said, there are a few issues with that one. The most important issue is we are always behind the wave. Whenever an event is happening somewhere in the networks, and that's the biggest topic for us, events happening somewhere where people are gathering, where you have a lot of people coming together, because that is totally ruining a static planning. That's why we move from a static approach into a more dynamic approach.
The second one is, whenever we look into analysis of topics, because you're right now in the physical network, you have to understand the issue, you have to understand all the parameters, and you have to make sure that is bringing the customer experience to the next level, so you have to get out of that one. Third one, pure sizing or pure scaling. One of the big issues we also face is that we have about 1,000 events which we can do manually. So 1,000 event where we can do all the analysis, running through all the details. The big issue, that's not reflecting reality. Reality is much bigger. The way we overcome that one is right now to bring the network planning, but also the reaction in the network to the next level. How we do that is, we built a team out of agents.
We built three agents, the one having a scanning function. Basically trying to understand what's going on out there, reading public media, social media, reading all the elements on the market, and trying to bring that one into a structured format. The second agent is the monitoring agent right now. He's taking over this information and trying to categorize the information. Putting T-shirt sizes on it, trying to understand what is the technical implication of that one, and how does this impact our network, and then as a consequence out of that one, customer experience. Last but not least, probably the most important one of the three, the remediation agent.
This agent is right now taking the information and trying fully autonomously to react on the demand in front of the event, already trying to prepare the network in a way that whenever the event is happening, we are prepared too, and customers have a perfect experience, and we can make sure they run in the best possible way. That's right now theory, and that's right now PowerPoint. I would right now like to switch between the theory to the reality, and you will see we already have that one in our networks. Probably do I have to switch? Ah, it's here. Sorry. I see now. Let me guide you through the process. Again, this is nothing which is made up for you here. This is how we work at the moment with RAN Guardian in our networks.
You find again on the left-hand side, the three agents. Let me just activate the first agent. What this agent is now doing is he is reading through the media set, trying to understand all kinds of events, sport events, festivals, community gatherings, other things, so to say. We are pulling everything together here, trying to understand what it is. You see there is a huge list of event populating down there. You also see it from a geographical point of view. The important thing is we are trying to understand what are the big events in there. What are the events plus 1,000 people which are happening in Germany in that case? You already see it's a huge number. This is a year's number. That's a real number.
It's 213K events which are impacting in any way, so to say, our network, which is huge. We take right now all this information, so whatever we have learned out of this agent, we put that one right now into the monitoring agent. When I activate the monitoring agent, you see a few very important things. You have, again, the events where it happens, the T-shirt size is set. Here is the first additional information. You see it broken down to physics. You see the physical antennas, the physical sites on this side here. The next important information is the cell status. Why? Because some cells are in a perfect shape, so we don't have to deal with it. We also have cells which need some action up front before we start the automation.
You see the amber ones, you better send a technician there, make sure, so to say, this cell also goes into the green state before we start the remediation action. Here on the right-hand side, you see what is the first estimation of the agent according to the size of the event, whether we can mitigate the risk and whether we can run, so to say, in an autonomous way. You see a few of them are sufficient, so nothing to be done there. All fine. You also see a few at risk. These are the ones where we will then see the autonomous agent taking remediation actions to make sure we're going to solve it.
We also see some red ones where as an estimation, we need some additional measures, which would mean we have to bring some additional mobile sites, we have to bring some more capacity to the site. This can't be dealt with out of the system itself. When I now go to the remediation agent, you see it's also populating right now. There's a lot of activities running. By the way, this is also just drawn from the system. It's real data you're seeing here. You see, for example, for a specific event, 24 action were executed. There are totally different kind of things like we change the carrier aggregation, we change the uplink, we make sure, so to say, the tilt is the right one. We really focus capacity to the place where the event happens. Then you see the outcome of that one.
Most of the time, we fortunately already see congestion could be mitigated automatically, so full autonomous activity. Some of them need still some human interaction to make sure we are going to solve it. I break it down one more level just to show what is in there. You see right now the sites. This is right now for this one event, all the touched sites and all the activities we executed. You saw, for example, or see here, we deactivated carrier aggregation. We spread the capacity more to the people, less high-speed capacity or high-speed service delivered to the customers. Less capacity, but spread to everyone. So more distribution of capacity.
And with that one, you can also follow all the steps which were done in the remediation device or which are done in the remediation agent to make sure in the end, so to say, you are always on top what is happening in the network, because that is the most important thing. We always have to stay on top what is happening in our networks. That said, this is the kind of interplay between the three agents. That is what is deployed in the network. That is also what helps us to get us faster, not just reactive, but proactive into the process in capacity planning and customer experience. And with that one, let me come back to the outcome of the whole activity. Here we go.
Okay. Here we go. The one is what Tim showed at the very beginning. We are really speeding up here. Why? Because it is all done within the agents. The whole identifying, classifying, remediation is done within the pairs of agent. The second one is the scale. I said before, we can cover about 1,000 events a year with human beings. Now we do 100 fold. We do 100K events, really analyzing, and really breaking it down and also remediate. And probably the most important one is we bring one thing to our customers, which is the avoidance of 24 million minutes of bad customer experience. That is the most and most critical impact we are bringing to the game. With that one, I want to conclude and probably do this with a short video.
Thank you so much. Arash, now over to you. Thank you.
Thank you, Armin. RAN Guardian is a great example how far we already have progressed in our automation journey. By the way, RAN Guardian is live in our German market since January. It will go live this year in Czech Republic, and multiple other markets will follow next year. As a leading telco, we are well advanced in our automation journey and have many specific use cases already implemented across our network. However, the ambition goes far beyond individual automation. As already mentioned by Tim and Rodrigo, we want to achieve Level 4 autonomous networks, and there are some challenges to overcome. Most of today's automation is still domain-specific and addresses specific use cases. At the same time, our networks are permanently evolving with new features, capabilities, but also increased complexity. Point solutions will not get us to autonomous network.
Agentic AI will be a game changer because it allows us to move from individual automation use cases towards scalable autonomy. Let me introduce MINDR, our multi-agentic AI platform. MINDR is a modular, autonomous observability and incident management platform that brings specialized agents across the network together. They can anticipate issues, reason different systems, and increasingly take autonomous action. In other words, MINDR will give us a platform to scale from individual AI use cases to an increasingly autonomous network. Let's quickly see how it's actually working. MINDR has basically four steps. The first step is it correlates all sorts of signals event across the network. The root cause analytics agents does reasoning with multiple specialized agents per network domain and network node to diagnose the issue. The third step is the remediation, which is the platform can execute and verify remediation within predefined guardrails.
Some of the issues can be fixed, and the impact indirectly is low, so we may choose that the agent does it autonomously. For every critical change, there will be always a human in the loop, which has to approve the change before it gets executed. Very important, the fourth step, which is MINDR is continuously learning from the outcome, and it gets better, faster, and more effectively solving the problem. Now let me show you the magic live. Let's start with alarms. I pick up one of the most complex service chains, which is Voice over LTE, which is your 4G native voice calls. You see just in the last 24 hours, there are roughly 20,000 or 30,000 alarms. No 20. A lot of them are gracefully closed.
These are minor alarms, temporary change, some counter work, and with a simple rule, you basically neglect them, and you close them. However, 52% of the remaining ones, we have today already good automation to take care of it. Around 1,400, which are more the complex one, requires human intervention. With MINDR, we can address a big part of that, and there will be very specific cases where, as of today, the agent does not have the solution, and it would escalate this into a NOC engineer. Just you, from roughly 20,000, ultimately only 48 requires real human activation beyond approval process. Now, we are picking up one alarm. Normally, MINDR would run this all autonomously, but for these demo purposes, we introduce a step-by-step function so you can follow what's happening. Gy is an interface for prepaid.
This looks like a prepaid issue, relate to Voice over LTE. It's in Frankfurt, and 4,200 customers are affected. Now let's start. Basically what's happening is the agents look at which domains are affected, correlates all information, even events in the network building up to that issue, figuring out the customer impact, the blast radius, which networks nodes are involved, what related alarms do exist, and have a good understanding of what the issue is. Next step would be the reasoning and correlations. Basically, the context agents look at this. They run health checks on each different node to see where the issue could be. Then the knowledge graph of how the flow works will be updated and permanently verified. Basically, there is an assumption what the issue is.
It will be validated until it reaches a level of confidence level, runs performance engines, traces everything automatically, and it comes to a conclusion with a 96% confidence level of what the issue is.
What is it?
I can explain in the Q&A if we have time. You see with the resolution, they say this change is a level 3 autonomy, so our guardrails require a human to approve. Basically, their resolution and even a proposed time window is addressed, and it says, "Look, you might do this tomorrow, or you act now for faster recovery." Once the NOC engineer approves, it's automatically execute that change. If it requires to dispatch someone, which is, in this case, not the case, it would generate automatically a ticket for someone actively there. More importantly, once the change has been implemented, it autonomously validate all the hypothesis and checks again the flow if the change was successful and if the desired outcome was what was agreed. As I said, one of the most important things is the platform learns.
Basically, it takes all the input and output into its learning system so that next time similar incidents can be done faster and more efficiently. Good. What it does is it will leave an interaction for the NOC engineer at any point in time to come back to say, "Was there reasoning?" Because this was a change to be approved by the engineer. "Did you get good information? How was the issue?" You can have a small survey, or it has a full chat interface which also the NOC engineer can chat with the agent to make it better. Again, normally this was run much faster autonomously. We slow it down and put a step-by-step so that people can follow. Now, what you have seen was just one autonomous incident group. One. Imagine this at scale in our networks.
MINDR is the key foundation for journey towards Level 4 autonomous networks. Together with our broader network transformation, we are targeting 90% of tickets greater than 1,900 to handled automatically. We are confident that we can qualifying the issues much faster, below 15 minutes on average, that incident resolutions go below one hour, and ultimately, because of the nature of our networks, to gain productivity of 30%. Productivity can mean that we can handle more things or that we move network experts to other parts. Generally, what this whole thing does with the human beings, that the general network engineer, instead of checking all the stuff and doing things manually, can go into high-value task. Autonomy cannot be just about efficiency. Ultimately, it has to translate into a better customer experience.
That's why we are very confident to introduce our new KPI with a new ambition, which is our CXI index, which is already above 3.5, to move it above four for more than 90% of our customers. With that, I'd like to show you a quick video what MINDR will actually be for our customers.
[Presentation]
As Rodrigo already said, we will go with our first use case as an MVP live by the end of the year in Germany, and next year, we will scale the platform with its full scope across Germany and many of our other markets. With that, I hand over back to Rodrigo.
Thank you. Hannes, you can come, because we now will have our first Q&A. To summarize, we showed you the story, a story of transformation, a vision on how we believe the network will fundamentally change in the future through AI, but also how AI will allow us to fundamentally transform how we operate those networks. And we showed you some cases how this is a reality as of today. Some of these use cases are live in our network. And this will translate into significantly higher levels of automation. Our target for 2030 is Level 4, which is just short of Level 5, which is a fully 100% autonomous network. So that means an autonomous network where humans intervene by exception. Our objective in terms of customer experience on the CXI, this translates to a nearly perfect, flawless customer experience for more than 90% of our customers.
And when I mean flawless, above four means nearly flawless. And finally, of course, we want to transform all of this into efficiencies only in fiber. This means for next year, EUR 100 million to EUR 150 million of efficiencies that are baked in into our plans. And I believe, as Tim says, when we set ourselves targets at Deutsche Telekom and share them with you, we bake them into our plan. But I personally believe that the potential of all of these technologies is much higher. This is what we can see today, and I assure you, we will keep setting the ambitions higher and keep thriving for more. With that, Hannes, the Q&A.
Okay. Well, thank you, Rodrigo. Thank you, team. Great stuff. So maybe, team, can you come back on the stage here for the network-related Q&A? So any questions from the audience on what you have seen? Polo, start with you.
Hi, it is Polo Tang from UBS. You talked a lot about the use of AI agents in terms of saving costs, but obviously one very topical question at the moment is the impact of AI agents in terms of your revenues and customers renegotiating their bills. So just wondering, Rodrigo, if you had any views on that.
We have lots of views on that, but actually it is mainly a topic because next section is AI and customer interaction, so it would be more appropriate there. But of course, Rodrigo is happy to-
Yes.
Talk about this from a point of view of.
Yes, I think in the next section, we will cover that quite in depth indeed. But we are very excited about that. I can share with you a lot of the examples of what we are doing already in Germany, where significant number of our customer interactions are already managed by AI. By the way, we see, for example, in AI chat that the resolution rate of the AI is already at human level. On voice, we are not yet there, but getting there. We are also increasingly using AI to drive monetization, to manage churn, particularly in this German highly competitive market, using AI to improve churn is a key use case. I do believe that building agents going forward is going to be less and less difficult as the foundation models evolve and as the foundation models become available and as alternatives become available.
This is something that we will share in the next presentation, but where I also believe that we have a right to play.
Yeah. Next question is from Carl. You must even grab the microphone.
Thanks, Hannes. Carl Murdoch-Smith from Citi. Just on your point about the improving mobile CXI and kind of getting that to over 90% of customers. In terms of all the improvements that you are making, and it is great that lower network fix results times, all this kind of stuff. How much of this is Deutsche Telekom specific and how much it is kind of versus just the industry improving? On that improvement, are you tracking what is happening for your competitors as well? I suppose as the quality leader today, is there more improvement to be made for maybe the other operators in the market and does Telekom just become seen as a utility? Thanks.
Well, if anything, what we are seeing is that network differentiation is becoming an even stronger point for us to drive growth and differentiation in the market. I believe that everything we showed you will further strengthen that, because I do not think that at any scale you can afford to develop and run systems like this one. So I think the same way that we saw on other systems and platforms that scale matters, I think scale will matter here when it comes to tracking. So the CXI is an internal KPI, but when it comes to tracking our network experience versus our competition, of course, we do that as well. We do that through other measurements. If anything, I would just say that we are in a very healthy place today.
Also what is interesting is that if you look at the world of agents and how agents make recommendations, it is not only about value, it is also about differentiation and network. So we actually see the world of agents as an opportunity to keep further communicating differentiation to our customers.
Maybe from you and the team as well.
Everyone wants to answer that one.
We will not fight for that one. Just one element on that one, because it is an important question, how we compare to the competition here. We take whatever the industry offers, that is our base kind of level, and we start from there. Whatever you have seen right now is on top to what the industry offers from basic kind of functionality. We put on top this extra insight and this extra kind of intelligence, trying to get beyond what the industry is offering. That has multiple flavors. You see a lot of different things here, but as I said before, CXI was developed internally, and I think Alex can give much more insight on that one.
MINDR was developed in-house also, RAN Guardian was developed in-house, which is really helping to leapfrog the industry and not just build upon the offerings.
Maybe I can add on a little bit. What industry is giving us is very important and crucial. To see it from the system view, from the technological point of view, what was missing in the whole value chain was the customer view.
What we did, what we developed internally and now across the entire footprint of Deutsche Telekom is we used the customer perception of the network from all angles you can imagine. It is congestion, upload, download, round trip time, how they are moving. We are not using CXI only in our mobile network. We are using it also for the home network and also for our TV. It is a combination of everything. This gives us a holistic view of what is going on on the customer side, how the journey is perceived from the customer side, and then we steer the investment technological-wise. We steer also how our commercial teams are approaching the customers. This is something very new. No one else on the market has that one. We are very proud of that and it is working quite well.
That is the reason, as Rodrigo was highlighting it, we won the most consistent network in Europe for the first time in our history.
Okay. Well, thanks, guys. Next is Mathieu, I think.
Hi, Mathieu Robilliard from Barclays. Thank you. A question that is not entirely on the topics you touched, but since you guys live and breathe with the network, I wanted to understand is if you are already seeing some signs of change in the pattern of the traffic. Is that something that requires more investment? Is that something that can be something where you can differentiate? I think you were talking about network sensing things. I am not sure I understand exactly what it means, but I get the picture. Or is it an opportunity to generate more revenue? So really how AI is changing the traffic from your customer point of view and what you can do with that?
So reality is, and I checked that Friday as a preparation, today, if you look at our network, specific AI traffic is just 1% of the traffic. But what is changing is the ratio between downlink and uplink. That is why what we are building, which is the only fixed technology which is symmetrical, is going to be a huge differentiator compared to any cable or any type of other fixed. Obviously, for mobile, Tim covered that we will see a new smartphone moment. We will have new AI devices. So ultimately, these AI devices will have memory and will follow you through life. So the use case, the traffic pattern in the network will change, which is we will design and adopt our networks to be more balanced between downlink and uplink. And again, all of these, Birgit mentioned skills.
All of the things what you see is built in-house with our best engineers. So we are able to see patterns, to anticipate, and make sure that we use AI to do these things in the future automatically and dynamically. Because what we do not want is you will have different use cases from B2B to B2C. We will still have for years a big chunk of our customers on smartphones, and we will not design a network specifically for this. We will build an autonomous continuum of compute and connectivity to serve each customer like the application and like what is being required.
I think next up is James actually has the mic.
Yes. Thank you. James Ratzer from New Street. I hope what I am going to ask is an allowable question for today, but I was very interested in your RAN Guardian presentation for solving for capacity crunches, and it seemed like the number one solution you always had was to deactivate carrier aggregation. Why does that help to improve network capacity? I always thought that actually increased network capacity.
Yeah. Should we-
Should I? Yeah, sure. I can. First of all, thanks for the question. It is a rather technical one. By the way, the observation is not fully true. There was a much longer list down there. I was just not showing all the elements which are up there. As said, we also change uplink and other things in the network. But in principle, what we are trying to do is to get at least a fair share to the people. Actually, there are two elements in there. The one is, we offer that in at least one of the markets already, is to give a share for a specific service in the network, so-called slicing. That is one of the elements where we do to secure certain services to run even in congested areas.
The second one is we try to get a fair share for the rest of the capacity to the people. By really using all the levers in the network through all the parameters, and as said, it is a rather long list on that one. It is, by the way, also tilting. It is bringing some neighboring sites here to serve the one where it is more congested. These are all elements which are helping to offload or to make sure there is at least this minimum acceptable capacity for each and every user in the footprint.
A cell site is like an umbrella, and the signal aggregation is like a Christmas tree, which is your lower part is the connectivity or reach layer, and the higher you go with the spectrum, it becomes more capacity. So when you have an event, what you want is to give as much as for people minimum requirement to have a good experience without having maybe the top speeds. That's why you lower down your carrier aggregation to have much more bandwidth at the bottom of the Christmas tree, if that makes sense.
Yeah. Okay. Okay, Josh is next.
Hi there. It's Josh Mills at BNP Paribas. Two questions from me. One, are there any examples so far with the autonomous AI network control where it's gone wrong? If so, how easy is that to fix? Then secondly, I think at the beginning, Rodrigo, you were talking about the importance of having AI RAN and AI close to the customer. What would that actually improve in practical terms, and how do you plan to monetize it? Is it that it will be easier for me to use AI and I'll get a quicker service if I'm a DT customer? Or is there a way of monetizing it through the enterprise side as well? Thanks.
So maybe that's exactly what I said, MVP, because we are hardening the system. The good news is the way we build these agents is they are agents, and each set of agents have an agent observer on top of it. Number one. Number two is never has an agent direct access to the network element. There is what's so-called MCP, which is a gateway, and there is defined agent I, agent A can do A, B, C, D, and only A, B, C, D with agent B. So like a firewall. Firewall sounds negative, but it's a strong control point. So that's the first thing. The second thing is, obviously, I showed you in the MINDR the network nodes agents, which is there is a permanent link between is there a misbehavior from a control agent, which is outside of the loop, just checking permanently.
And obviously, wherever, we have rarely seen this so far, to be quite honest. In the testing, sometimes it's not misbehaving, sometimes it's behaving differently than what we expected, and that's why we are hardening it. The platform is quite advanced, but giving a critical network, we just go with one use case because we want hard on that. On AI for network, you want to take it?
Yeah. Just the monetization part, I will leave for Rodrigo, but let me explain you the technical part of it. Just because we had this question, also a similar one before, but to give you a real-life use case.
What is important for tomorrow's network is the upload, not anymore the download as much, and the latency. Rodrigo was explaining that one. Why? I give you very practical example. Imagine you have physical AI. We talk about physical AI. Imagine you have a robot. Imagine it would take very long when the robot sends something, sees something, tries to process something, and it takes very long until it will send somewhere to a data center or to an edge computing near the network. That's where we play now the role. You cannot send this data going overseas to U.S. and be processed there and comes back. Of course, it can be done in a faster way, but still, the latency is too long. That's where you cannot utilize that one.
That's where we play a role, where it is very important to have also the AI run near the customers, near the use cases in order then also to be able to later monetize it.
Yeah. So.
Yeah, I think that's why I use this analogy, which to me is very accurate, which is think of the cloud as the brain and think of the network as the sensory or the nervous system, and the network will have intelligence. So what the network will see is different AI workloads running on the network. And with that intelligence, it can optimize which are the workloads that can, let's say, or need to go to a faraway data center with the jitter and the latency implications of that, and what are workloads that can run much closer to the end user to make that particularly robotics or whatever AI use case work. So, that's also what we call orchestration, the network that kind of routes this traffic based on intent and based on need. I believe, yes, that this will open avenues for monetization.
For sure, if you think about defense, if you think about public. By the way, some of these things are starting to emerge today, like for example, drone detection and these type of things that we do today in the network. So think about all of the use cases that will come. If you think about robots, the reality is the networks of today are not yet ready in terms of latency, jitter, et cetera, to have these robots running around in the open. So yes, I believe there will be opportunities for monetization. There will be opportunities for monetizing specific quality assurance, specific outcomes. I think it will be a monetization that is going to be very different to the ones we do today, which is a package of data. I think we will move more and more to monetizing specific outcomes based on the need.
Very good. I think Ulrich is the last question for this session because then you have a lunch coming up. So Ulrich, the pressure is on you.
Really, yes. Thank you. Ulrich Rathe from Bernstein. This seems a German-led technology development, and I think you alluded to rolling part of it out elsewhere in Europe. So it sounds like the corporate center, the domestic business developing this for Europe almost. Could you confirm that's the case? In this context, talk about to what extent you have worked together with the U.S. because obviously that's one of the question in investors' minds, to what extent the current group structure influences your ability to work together also on a technical basis and realize synergies there. Thank you.
Yeah. Okay. Let me probably start because parts of my job is to make sure we can scale in the IT space. For us, the principle was always the same. We build once, we deploy many times, and we make sure, so to say, the solution we created, and by the way, it is not important where it is created. So the example we have shown is a German-related one. The CXI was a development within the whole group, so there is much more to tell about that one. But what we will make sure is we have the same capability built on the same agent structure, built on the same underlying infrastructure across the group. So we are working on the same data structure, the same way we store data, and really also how we expose it and how we run the agents on top.
That will be out of one mold across the full footprint.
Okay.
For the U.S. piece-
Yeah. I think we can add some-
My team is building all these capabilities. Armin will scale them. We have built an AI marketplace with use cases, and there are a lot of amazing use cases coming from Greece, coming from Hungary. We just chose here some specific ones, so the impression that is all German-led. We have a big, large organization here. We do many things here, but there is a lot of, I think, good, and this is what a group is about. In that marketplace, we have, I think, 50 use cases, many of them. Some of them are not even from center, is from the great work our other markets do. We validate that, and we make it available, and Armin will then scale it.
On U.S., Tim mentioned something important, and me and part of my leadership team spent a good day, actually, with John Saw's team a week ago, is for 6G. Everything toward 6G, we are together going to lead in the West world. We are collaborating heavily on standardization to avoid, remember in the past, NSA, SA, all this, to have one standard. Why? Because it is going to reduce the variations and the R&D cost of our suppliers on our R&D power is not what it used to be. We have announced at Mobile World Congress, and you will see very soon much more detail of that, a joint innovation hub Berlin Bellevue, where all the things from physical AI sensing, intent-based networking, we will develop together and we share.
Which is they take care about the specific use case, share with us the results, we will do the stuff Europe. I fundamentally believe 6G and the path to 6G is the best way of collaborating beyond what has happened today.
Maybe also adding to this, when you look at some of the demos or use cases.
Yes.
That you see presented today, we have the RAN Guardian and T-Mobile in the U.S.A. They have a different name, it is Dynamic CX, but it is the same concept and each one of us is the first operators in the world to have this. Clearly there is an exchange of innovation, of ideas, and then best practices as well. I think Kartik wanted to add something.
I think the rationalization across countries and having one layer is also true for the UDP, the data layer that we spoke about.
Yes.
That is a common layer that is a reference architecture to which all of the future data platforms will migrate to. As we build new stuff, all of that is being built harmonized across all countries.
Okay, excellent.
Maybe just one note, because you said this is German-driven. I have worked in several positions within Deutsche Telekom, and I can tell you one thing which we are all in Deutsche Telekom proud of is copy with pride within the group. If one country is doing it good and this is copied to the other countries, and when it works, when it is proven that it works, then we are going to do that. So it is regardless if it is coming from Germany or somewhere else. So the collaboration and the cooperation between the NatCos is a power we have on the market.
That is why we brought you from Poland, actually.
Okay, very good. Thank you all very much. In terms of organization, we now have lunch, and the lunch, I think we have 30 minutes of lunch. I am being told. No, it says different on my instructions because we are running a bit late. I was trying to use the lunch to recover a bit of time, but please be back here at, let us say, 2:35, and from 2:35 to 2:40.
24 minutes.
Yeah. Okay. Latest 2:40. Thank you, Kartik.
Welcome back. I hope you enjoyed the food and the conversations. We continue the AI Investor Day with our second business-related session, AI and Customer Interactions. Dominique Leroy, our Head of Europe, will lead through the session, starting with our strategic thinking, our AI will transform customer interactions before introducing, again, three demo cases that showcase the contribution of AI towards this vision. Dominique, over to you.
Good afternoon, everybody. Sorry to come just after lunch. I still think I will be able to entertain you because we are talking about AI in the customer interaction, and that is probably one of the domain that you know most, and you probably have also some questions. Let me start and introduce first the team that will give you the demo. We have Lena Drubel, who will present you AI calling. Then we have Peter Meyer van Esch, who will present you everything we do in Germany around customer service. Then we have a nice duo with Uli Klenke and Lilla Kovács that will present you in what we do on brands and on marketing activities with AI. So why is AI so relevant in customer service? Because you see the figures on the left side of the chart.
We have 100 billion calls that are happening on our network every year, 100 million contact center interaction per year, 70 million shop visits, and 200 million of our services that are already managed through our app or in a digital format. So that brings a lot of contacts that can be, to a certain level, automated and improved through AI. For me, it is not only about efficiency, because very often we talk AI will generate efficiency in call centers and shop and others, but it is also about customer experience. I think AI is a tool that we can very much use to enhance customer experience as well as bringing efficiencies. We have three parts, one product, one sales and service, one marketing. Let me start with the product part. You already heard in the last sessions that we have networks that are becoming more and more intelligent.
Thanks to that, we are currently using a lot of this intelligence in the network to provide new services to our customers and hopefully monetize them. That is, still anticipating your questions, still something we are working on, and we will see in function of the adoption. The good thing on having intelligence in the network and using AI is that you can bring smart, secure, and device and app agnostic features. Let me go a bit more in detail. The AI calling is something that will be launched in Germany and in Greece, will do three things. They will be able to do live translation. So, for instance, if I talk to Tim, I can talk French, he can talk German, and the system translates. So I will hear Tim talking French to me, and he will hear me talking German to him.
Call summary, also a very good use case, very often also for B2B. When you do a call, you hang up, you want to have some summary of the call. This is something our AI assistant in the network can do for you. The last one, which is probably the most challenging, is when you call, you want to ask something to either your calendar or to the web or whoever, you can also do it directly in the call like Lena will present to you. A good benefit of that is that it is this unique call experience, but it is also very much quite natural interaction because we use voice-enabled AI, and so everything happens through a voice answering to you. As I said, it is native integration in every call.
It operates across devices, so if I am an iOS customer and someone in the room is Android customer, that system works because it works on the network underneath. You do not need to have the same type of devices or the same type of operating system. Today, you know how difficult it is to communicate from an Apple to a Samsung device. This here is totally solved because it is based on our core network. Also from a security point of view, as it is baked on the network, it is security by design. You do not need application and layer on top of it. That is the first product. I stop here because I think you will hear much more from Lena in the demo. The second part of the presentation is everything we do in the sales and service area.
More and more customer service will go through an AI-infused voice or bot. Still, we want to do that for automation, but of course, we will still use human for a lot of other interaction. The first example is something that I think most of the telco are doing, which is AI bot and AI voice. You can reach us 24 hours a day, seven days a week. We are able to answer you. The bot is very mature. The voice is getting pretty mature, where you can interact with us in different languages, with enterprise-grade security, and we will answer you all the question through the different channel that we interact with you, being call, being app, being web, or even on social media. One example that I brought here, in Greece, we have launched a voice bot relatively recently with the help of Wonderful and ElevenLabs.
In only five weeks, we were able on the TV interaction, so it is not on all product, but it was very much targeted at all the questions we had from customers on our MagentaTV. We have 49% resolution rate in five weeks, 30% faster than human, and 92% accuracy on answer. So pretty impressive figures that we are now. This is from few months ago. Now we are scaling that to many, many more products. The second use case in the sales and marketing is making our call center people much more efficient. We use AI before the intervention. As soon as someone contacts the call center, we often are able to identify the person through the mobile number. We can prepare the customer profile, and we can bring to the agent a summary of all the interaction we have had with the customer before.
During the interaction, the call center person can get a lot of support from the AI, being intent, context, knowledge, grounded responses, everything we can ask the agent to propose to our customers. The human agents, sales, retention, different type of action, that is all prepared by the AI. After, it is also increasing very much both the speed and mainly the accuracy of the filing, either a case summary or direct actions that the AI can do. A few examples in terms of figures. Today, we are working on this with Sprinklr. We have more than 12,000 agents that are live on Sprinklr today in Germany and in Europe. We have transactional NPS for contact center that are above 67, and we don't see degradation while we are using bots.
Call transaction time is decreasing, and certainly for complex calls, the time to resolve complex case is really lower. We also use AI in the shops. In the shops, we have here three examples. The first one is MIA, who is an AI digital human, as we call it, where in the flagship store, we don't have that everywhere because it's still relatively expensive. She greets the customer, she can start to answer a few basic questions, and the purpose is that she would be able to answer more and more questions going forward. That's one use case. Second use case is how we train the shop people.
We have learned in all our country, by the way, that being trained by an AI is much more efficient for our shop agents than it is by a supervisory person or a shop manager because they dare much more. They don't feel judged or assessed by the AI. We see a lot of improvement in terms of quality of the answer, but also a lot of reduced time to onboard new shop people in our shop, thanks to the usage of AI in coaching. The last example is also something we use during the interaction with the permission of the customer always. When the AI is listening to the interaction in the call center, it's logical, but in the shop, we do it more and more.
We have the AI giving direct feedback to our salespeople so that they can really improve also on the spot, the way they talk to our customers. Last example on this section, the sales and services, before I go to marketing, is how we use AI not to solve interaction, but to reduce the number of interaction because we do preventive calling and preventive interaction with the customer. Few example, the first one from Germany, where you hear that we do a lot of new fiber provisioning. You saw how we try to improve very much the laying of fiber here. We try to improve once a customer has ordered a fiber till it is fully activated. We follow that process through an agent, and we have either direct communication or even AI outbound call to make sure that the customer feel very much taken care of.
Through that, we have reduced by 90,000 the troubleshooting ticket, and we have 30% less provisioning issues. Another example from Croatia, which is about the bills. We very often know that when the bill is quite different from your previous months because you have been abroad or whatever has happened, we call that bill shock. A customer, first thing he does is calling us or going to a shop to complain about the bill. We can also anticipate that. We can see in our system before we send the bill that the bill is a bit of a different type of bill. We can then proactively talk to the customer through AI to explain what is the reason for this higher bill. Therefore, we have way less customer calling us and also more happy customers because they understand why the bill is potentially higher.
Other examples, in Germany, a lot of things that you need to fill in or document we receive is scanned through AI, which improve the efficiency, so 8,000, 100% automation rate increase in the B2C. Also in B2B, this is something we are testing currently in Croatia, where for small customers, VSEs or very small customers, we currently have 15,000 accounts that are fully managed by AI. Currently, both from a pipeline management, a sales, and a care is fully done through agent. We will need, if it works, we will then scale it, but for the time being, we do a test on 15,000 very small enterprise, which is already quite significant. Let me jump to marketing and the whole hyper-personalization. AI is also a huge help to have much more contextualized and personalized offer to our customer. We say here individually created.
This is something that is in progress, but this is really the objective, is to really know the customer so well. We have so many contact points through our network or through Magenta Moments or through the interaction that the people have with us, that we have a very accurate knowledge on who are our customers. By using them, we can personalize much more the offer we give to the customer. Through AI, we can do that in a much more granular way than we did it so far. The same is also true for content generation, and Uli will present that we are able to do brand campaign, but also promotional campaigns for personalization with AI-generated content. That's also allowing us to be much more present and accurate in terms of attracting the customer attention.
That's where you get this time of high figures, up to 10 times higher sales. For instance, this was a Wi-Fi extension by targeting people in some circumstances. We had a very high conversion rate also in Slovakia, conversion rate on a much better household churn prediction. In Germany, the same much more targeted retention offer with 1.7x retention. On the brand campaigns, we are seeing that when we use AI, we have a decrease up to 30% of the cost of content creation. I think that's short, not so short, because I'm over time already for quite some time. We will start with the Magenta AI Call Assistant, with Lena, but before, we will show you a small video to explain to you what is the AI calling. Thank you.
[Presentation]
[Presentation]
[Presentation]
[Presentation]
All right. You heard how exciting my colleagues across go-to-market network product about launching it this year and sharing about the AI calling experiences that we are getting to the first hands of customers this year with the three use cases, real-time translation, in-call summary, and getting access to world knowledge right in the phone call. I personally have to say, as someone being in product, as much as I am excited about what we are bringing, I am super excited about how AI is actually changing the pace and the way we are developing those products. When you think about we just built this product as a prototype within 8 weeks, fully functional right in the network.
When you think in the past, I would say work like this, when we build a product, really completely new product, carrier-grade product, completely from scratch, this could have taken us three years or even more. This time, we went from planning to production in less than 12 months.
One of the reasons on how we got there was that we built a flexible architecture right from the start. This gave us the opportunity to work with multiple partners and to also co-build with them, to learn with them, and to iterate with them on the products as we are building them. Always adapt to the capabilities that AI is offering at this moment. When a new model is released or when a model's capability crosses the bar of actually what we require for the experience, we can just push it to production. Another reason that we have done this is we were able to build this once. The same product, the same architecture that we are building, we are bringing it to Greece and Germany this year, and we will further scale it within the footprint following next year.
The reason why I am actually sharing so much about the how is that this foundation that we are building really matters for what actually comes next. It really matters beyond the launch that we are doing this year. Because think of what we have now built is really voice as the interface for our customers to interact with. We have built this assistant that is sitting in the phone call, but that we can seamlessly also transfer into further touch points, into products, interactions that our customers are having with us already today. We are meeting them actually with new AI services right there where they already love using it, using it every day across basically all their life and everyday life. I would say that is more or less already kind of a wrap on what I wanted to share first.
So maybe as a key message, what we are doing this year is the starting point, but everything that we have built is actually built to be the starting point for what comes next. Talking about what comes next, I thought it might be most valuable to give you a glimpse into what the assistant can do in the future, how it is supporting our customers in the future, and how it is creating value for them in these situations where they need it the most. Therefore, I actually asked my colleague Julia to give me a phone call. All right. Hi, Julia. Thanks for calling. I was wondering actually, whether our movie night is still on for tonight.
Absolutely. Any requests?
I would say so far I had a pretty calm day, so we can go for something exciting, but please, no horror.
I am not into horror these days either. Let me take care of the food for tonight, and I will check the fridge first. Oh, it is actually completely empty.
Seriously?
Unless you want mustard for dinner.
Tempting, but no, thanks.
Hey, Magenta. Let's put together a Deliveroo order.
I-
Sure, I will keep track.
Perfect. Then I definitely go for a salami pizza.
Sounds good. Any snacks?
Yeah, no movie night without the popcorn that I love that much.
You got it. How about something sparkling drink for tonight?
Yep, let's do that.
Got it. Salami pizza, Lena's usual popcorn, and alcohol-free sparkling drinks. I'll send the order to your phone after the call.
All right, perfect. Lena, what do you think? Which time can you get here?
Oh, that's a good one. I'm actually still in the office. Hey, Magenta, how long will it take to get from the office to Julia's place?
Take the U2 three stops, then it's a four-minute walk. You'll be there in about 20 minutes.
All right. Julia, I will try to catch the train.
Sounds good. See you then. Hi, Lena. Right on time.
Yes, right on time. I made it. Thanks for preparing all the food. Tell me, what are we going to watch tonight?
Hey, Magenta. Any recommendations for us tonight?
Picks for the two of you. Exciting, but no horror.
All right. That looks good. "Silence of Nora." Let's go for it.
All right, let's play it.
All right, and from the living room over to Peter.
Good afternoon, everybody. We are now moving from popcorn and salami pizza and AI products in our network to AI in customer service, sales, and service operations. Pretty happy to share something around what is going on in our service operations. First of all, you can imagine AI is currently already changing everything, how we run service operations. As of now, every customer interaction, every customer conversation that we have is already handled by AI or even assisted and supported by AI, and I will show you later. Let's put this into practice. We start always with some conversational AI. In our cases, this is Frag Magenta. It is our chatbot, it is our voice bot, and we are offering this along all the channels that we have. It is always about solving customer needs just before human capacity is needed.
Whenever human expertise is needed, such a bot handles a seamless, smooth transition to our agents. I walk you through a typical journey that we have a thousand of times every week. It is a new fiber customer, and I show you along the journey where AI is in place already. Let's start.
[Presentation]
Okay, in case the agent has to take over, the agent doesn't start at zero. The AI provides relevant customer information, a short AI briefing, where everything he or she needs to know is already in there. It's customer contacts, it's summary of the last contacts or customer history, and makes it much more easy to enter a new conversation with a customer. The customer feels like he's known as an individual to us, and it, of course, helps decreasing handling time. But on the other hand, push much on customer experience. That creates a trusty atmosphere where sales is much more possible. Better handling time, better customer experience, and we all show this in our 360-degree front end, we call it MagentaView, where you get everything you should know about the customer.
And we pack this front end with a lot of new AI tools, I would like to share one of them with you.
[Presentation]
Okay, this is interesting. During the conversation, AI listens to the call, and it identifies specific things that happen and occur along this conversation. This can be a sales opportunity, as in this case. The AI hears, listens, and finds out this is the right moment where we should provide a sales offering, because this is the right information at the right time for the agent to do a sales pitch. On the other hand, there are lots of care issues. You can imagine, something, a question about your bill, et cetera. And the AI does the research part and offers already the relevant information to the agent so that we do not have to research anymore. So handling time drops and, what is super important, it keeps a kind of a natural conversation flow along this journey.
Because you do not have to stop, go into research, offer some different systems or whatever. As an agent, you get the right information at the right time. So, in this case, where there is sales opportunity, this is a great combination where productivity gains turn into growth opportunity as well. And when the call ends, the AI still works and supports the agent. So I will show you.
[Presentation]
Auto-documentation. In the past, every agent was documenting after each call what happens. Nobody likes it, as you can imagine, and so was the quality in documentation. AI can do this job much, much better and much, much faster and much cheaper. This is not a pilot or a POC. We are doing this 50,000 times every day. Beside eliminating the work part here, there is one topic which is much more interesting, because this offers a lot of customer insights. Along a conversation, we get to know the customer better, and we can document a lot of new information around the customer. Every conversation creates new data, new knowledge that we can put into an insight engine, as we call it.
We take all our 100,000, millions of auto documentations and put it into a separate place, the AI insight engine, and we add everything else we know about the customer into this insight engine as well. Now it is interesting because machine learning enables us now to predict the next most probable step or action of the customer, sometimes even before the customer knows. This shifts a core idea of how service could be. We are moving from a reactive service, waiting for customers to call us, into a proactive, a predictive, proactive service where we know what the customer is going to do next, and we can react on that one. Let me give you an example of what happens when we connect all this, and what I show you is in place already this year.
We have implemented a model in our cases for new fiber customers. The difference compared to, some of you may have heard of churn prediction and churn avoidance programs. This is about order cancellation rates. Between an order, an incoming order, a customer that gives us an order to provide some access, there are sometimes just a couple of days left till he cancels his order because there are some issues we do not know. This machine is now able to identify a risk group where we know that this group, those customers, are likely to cancel their order after sales. What we do now is we can take action on this, and we start calling them. Just ask for, "Is there anything open? Anything we can do? Do you need some help?" We see multiple different reasons.
Sometimes just like, "I am not sure. I got an information from my provider," et cetera. Then we can help. How this feels, shows the next video, just to give you a short understanding.
[Presentation]
To be honest, we were surprised by the results. We started this initiative, and we were able to lower cancellation rate in this risk group by 50%. This is amazing. We were able to safeguard, based on this year, 12,000 customers, new broadband customers, avoiding cancellation. We do really tough and strict A/B testing all the time. To expand this, there are a couple of opportunities right now, and we see 50K roughly around potential just by scaling this single use case. If you just imagine this customer insight engine, this is such a great capability. Use cases, I would say, nearly endless. Shifting from reactive to proactive offers so much more. I'll move on because the next part was interesting as well. Those of you who have a feeling of outbound, call outbound business, know it, customers usually don't like it.
If they get called on outbound calls, usually it's something with sales. They don't like it. But in this case, proactive service, predictive-proactive service, taking care about our customers increased our journey NPS by 15%. It's making the whole journey better. Let me recap. The first thing I want to highlight is it's not a collection of a single AI use case that you've seen today. It's more like we are reinventing the whole customer service, the way we proceed, the way we think about proactive service. It's more that we start building customer service around AI capabilities. The second part, beside moving from reactive to proactive, is AI isn't just an efficiency game in customer service. It's an experience game. It's pushing customer experience. Of course, there's so much opportunity in sales and growth as well. Thanks for this.
We are now moving from AI in customer service to AI in marketing and brand. I am happy to introduce Uli and Lilla.
Thank you so much. Thank you, Peter. Lilla and I, we are going to walk you a little bit through a new kind of experience. We have a premiere for you today. Come to that in a second. AI is changing everything, what we do in brand and marketing. It is really interesting. We are trying to handle AI as a new kind of operating system. It is not just a number of tools we are using. Using AI in brand, it means we have to make the brand machine readable. We are working with the colleagues of Frontify to make it accessible to all LLMs. Whenever somebody is working on the brand somewhere, she or he is getting the right information in place. The first one is to make it accessible to everybody.
The next one is to use AI for content production in what we are doing all day. You saw that with Tim's avatar today. I got something else on the plate. This is a leadership thing. You have to teach the organization. You have to bring the right tools to the table, to work on the mental availability of AI in the people's heads in marketing and in the brand. You need to come from reactive to predictive. We are using a lot of data we are getting from the LLMs. We are reading the LLMs constantly. We are using the LLMs as a means of communication. Maybe you saw that we were one of the first users of ChatGPT Ads here in Europe.
On the other hand, we are trying to influence what is happening on the LLMs and what they are selling, what they are telling about us and about our brand, and how they service our customers. We are trying to produce content with the help of AI since a year, more or less. The first thing we did was the famous AI challenge. We disrupted the marketplace and put a radio script in the market and said, "Can anybody, for EUR 10,000, take part in a competition? Send us the best movies." We took these movies as a learning model. As you can imagine, many production firms went completely mental about this. It helped everybody to learn, get to the first steps. We are now in the fourth level. We have the fourth fully integrated campaign out there.
We are preparing our Christmas campaign. This is the premiere of our Christmas movie, which is completely made with the help of AI. Everything is artificial in there. We are getting better and better. As you can imagine, the team said, "No, we have to do that Christmas movie on camera, of course." With a director and real snow and interaction between people. We said, "No, let's try. Give it a try, to get into the fully integrated production of a Christmas movie." Here it is. You judge by yourself if you are getting emotionally connected or not. I can tell you, 50% of the emotional connection is the music. Here is our Christmas movie.
[Presentation]
A little bit too early in the year, but really interesting. We have an estimated price of EUR 1,000 a second. This is the price, more or less. You can imagine what an effort it would be if you would put that on camera. It is different now. I like AI best in content production when it expands our opportunities and possibilities like walking on these balloons. This is something you could not do in real life. This is just an example. There is a lot of tasks. We have more than 100 avatars, in sales and service right now in different kind of countries for different kind of purposes. Dominique was talking about MIA already. Magenta and MIA is very brand orientated. Of course, the naming is there. It is a digital human. We are trying to treat it this way.
It is really a lot of decisions to be made when you create an avatar. For example, one is the voice. Is the voice male, female, or neutral? I want it to be neutral. Then we tested it and found out, no, people do not want it if they do not know and cannot tell it is a man or woman. 60%, 70%, we are on female voices now. The rest of it on male voices. We do not use neutral voices because people are reacting negatively on this. This is MIA. I would like to. I am sorry. I tried to get back to MIA.
[Presentation]
Here it is.
[Presentation]
MIA is one example. We have a sound logo wizard like for the da-da-da-da-da. Wherever the music ends in a spot, it adapts, shifts a little bit up, or pitches a little bit down. Of course, one of our biggest brand is it is a magenta color, and we have millions of pictures out there. We have built an agent which automatically corrects the color code to the right magenta color. Whenever somebody in the organization is producing a photo, now uploading it here, and then the right color comes off, which is really valuable for us.
[Presentation]
What we do is, we ask the AI how our contents work like, so predictive marketing again. We send it over, and then we are getting a feedback. This is pretty close to what people say. We are comparing it to market research with real people. We are roundabout a predictive quality of 80%, but we are still doing everything in parallel, so it is hybrid market research what we are doing right now. But the prediction is getting better and better. Of course, we are one of the biggest spenders of media in the marketplace. Wherever we are in all the 10 countries in Europe and then in the U.S., we are spending a lot of money in media. For that reason, it makes sense to optimize this in planning, buying, and correcting, optimizing the data usage. It is good to have AI in place.
We are working with our partners from WPP right now. We are on pitch in that industry, and it is a really important decision for us. We estimate that we can save up to 25%-30% in agency cost in that field. Whenever we have some testimony in place like Thomas Müller for the World Cup campaign, of course, we can translate them, and he is pretending to speak French here.
[Non-English content]
So it's like this will be the next step, to avatarize testimonials. Super important because we don't need them for production. For the entire production of this World Cup campaign, we've been flying around the world. We've been to New Zealand and in Canada because he wasn't available. If we would have an avatarized version of him, we could use them in any production like a normal person. This will be the next step.
[Non-English content]
We are partnering with Black Forest Labs, and this is one of the very interesting partnerships because it is German. We have a stake in Black Forest Labs, and we are working together with them, mainly on the FLUX model. We are using it. We are alpha tester for the FLUX model. We are living that partnership. That means we've been creating a platform inside of Deutsche Telekom, which is called Create AI, where anybody can create photos with the help of the FLUX system, which is brand safe. We put in our own models there. We've been photographing people, avatarizing them, put them on the model, and the result is a number of people which are just like for us. Like in the ancient times, when there was Naomi Campbell and all the other super top models. We have this number of people.
We have contracts with them. We are allowed to make them younger but not older. This is very special about using people in the AI surroundings. Everybody around the globe in all marketing organizations and internally can work with these people. In the midst, here you got Anna, and Lilla was working with her and shows you now how it is in the real world.
Thank you so much. Hello, everyone. I am now going into more details how we leverage AI in our marketing campaign. As Dominique mentioned earlier, we are collecting a lot of data points, and from the data points, we are creating actionable customer insights. We are knowing much more about our customers. We are knowing their household context. We are knowing the Magenta Moments interaction. We are knowing the digital affinity. Now we are better target these customers. We are leveraging AI, like whom to target, what to offer to these customers, when to reach out, and how to deliver, what channel to deliver the message to. With all of this approach, we are delivering personalized campaigning to our customers, which will bring higher conversion rates and also higher value for our customers. What does it mean really in business value?
On three levels, first of all, it creates communication efficiency. As you could heard, with the Black Forest Labs AI-generated creatives, we can create tremendous amount different communications. Second, we can increase customer engagement. One of the good example that we have done that how could we engage Magenta Moments customers, like we did an AI segmentation and recommendation to our dormant Magenta Moments user base, and we reactivated more than business as usual, 30% higher share. Then the third one is the business value, what we can get, the business economics. As also Dominique mentioned, we had a campaign which we achieved 10x higher fixed upsell by selling the customer mesh products with the dialogue marketing, with the conversation building on engagement. Let me bring you this closer with an example.
We run a campaign of how we can upsell customers onto a higher TV package. We needed to find the right customers who we target. First, we did a household segmentation based on TV viewership data. Then we segmented the customer base who are movie lovers. But we did not stop there. We wanted to narrow down our target segments in order not to waste any communication. So we decided to offer to these segments, to this movie lover household segment, cinema tickets via our Magenta Moments platform, free cinema tickets. So we targeted those customers who redeemed these free cinema tickets. So our target base got narrowed. Then let us meet Anna. Anna is our target persona. Anna is a mother of two. She is an FMC customers having broadband, TV, and mobile subscription, and she loves Magenta Moments.
She redeemed tickets in the last campaign, but she is also loving movies and entertainment. With this richer information and clearer persona, we can really know not just only what product the customer using, but also the customer interest and intention and engagement with us. And we targeted Anna via the OneApp because of her digital affinity. So, what we put in front of her is an AI-generated offer, which was a CineStar premium pack to Anna. And when we reached out the customer in the right time, when she used the cinema ticket and when she went to the movie. As said, OneApp was our target channel, and we created literally for the persona who we talk to, a tailored visual and a tailored tonality and message. Anna was not the only one who we targeted within this communication. We also targeted different other personas.
With the help of the Black Forest Lab, we created different creatives with different tonality and different text. With this, we achieved quite good results. The customers who we targeted via SMS, we achieved 28% higher upsell ratio. The customers who we targeted via the OneApp, we see 129% higher upsell onto more than EUR 5, which is a super good result. Also on the Magenta Moments platform, with the household segmentation, we more than doubled, almost doubled our cinema ticket redemptions. All in all, with having the right customer insights in place and leveraging AI to give the better commercial decisions and commercial offering in front of the customers, we are leveraging AI to find the right customers, to select the right offer, and give it to the right channel.
We are learning all the time with the response what the customer is giving to us. This is how we scale personalization. This is how we create additional value to the customers and increase higher conversion rate in marketing. Now I ask Dominique back on the stage to give us a bit of summary.
Thank you, ladies and gentlemen. So very quickly, you saw three big examples on the AI calling, which is I think really a revolution using intelligence on our network and going across device and operating system. You saw with Peter quite some examples where we can scale efficiency and scale sales in the sales and service, and lately with Uli and Lilla, how we can leverage growth and do more advertising thanks to AI. This is then the slides on conclusion. What is then the outcome of all that? Because it's nice as we come with a lot of story, but I think you as investor, you are very much interested what's in it. I think these are our targets for 2030, but I think you have a lot of proof points that convince us that we can get there.
We have a target to reduce entering calls and chat by 50%-55%. We want to achieve best resolution rate of 80%, 90%. All the transactions that are linked to retention of tariff change that we can do via digital, we want to get the digital interaction up to 70%. So it means 70% of people retaining contract or changing tariff, we don't have any interactions anymore with human. Everything is done via digital channels with the help of AI, and we want to increase conversion rates of all our promotional offer by 40%-50%. So in terms of outcome, this should give us around 20%-30% saving in sales and service. This is before the token cost, but this is what we are aiming for.
We think we can further increase our transactional NPS by five points, and we are aiming a growth in the B2C area of 1% and 1.5%, which will be there through the use of AI. Half is coming from lower churn, and half is coming from more upsell. This is the growth AI generated, not the full growth. So I think we will grow more than that, but this is what we can get through AI generated. Yes. That's it, and I think I call Hannes for the Q&A.
Thank you, Dominique. Thanks, Lena, thanks, Peter, and Lilla, and Uli. Can you please all come on the stage for the final Q&A?
You stay in the middle.
No, you should be in the middle.
We can swap.
Okay, so I have a question here from Josh.
Thanks. It's Josh Mills from BNP Paribas, and thank you for the interesting demonstrations of what AI can do. I suppose a few questions from my side. Often when customers are reaching out to you proactively, it's to negotiate on bills, it's to make a complaint, and this is one of the things that investors are increasingly worried about, is that you'll see a wave of right-sizing bills across the industry. So the first question is, what percentage of your incoming contact with customers today are people looking to pay less money or to renegotiate their bills? Secondly, if your agent comes up against MUSE, Instinct, one of these other AI agents, how does that play out? Does it just keep renegotiating until you get the best retention offer, or do you have safeguards in place to stop that from being the case? Thanks.
Okay. Let me give a first answer to that. I think the agent-to-agent communication, because I think that's mainly the content what you see upcoming currently in the U.S. with MUSE or with Instinct, is something we will not stop. We will not do like Amazon, stopping the interaction. We think that there is also merit in having this AI to AI interaction. Today, in a lot of countries, we have also already a lot of tariff platforms that are comparing prices. Check24 in Germany is already very active, used by a lot of people. We think if we are able to transfer that platform into agent, we get a much more intelligent comparison of prices. Today, the comparison of Check24 is just blunt prices, one price versus another price.
If you give that to agents that are intelligent agent, you can bring in there much more sophistications. By the way, the only way for agents, external agents, to communicate with us will be at the moment that we will be exposing an agent from our customers onto the web. This is something we don't have today, but of course, we know what is coming, so we are preparing. In this agent, we are able to feed it with information from the person if the person give us consent to expose the information to the web, and that's a much richer information than anything you can grab from the web. So you can bring the information about the convergence. You can bring the network quality, the brand quality.
What we will need to do as well is to, today, we give a lot of elements that are in the offer, but not explicitly. We will need to bring the network quality in a way that is quantitative, that an agent can assess versus other network. We will need to make sure that we bring the advantage that we can have with Magenta Moments. If we have security-embedded product, we will need to make them explicit. I think this all new agent-to-agent will have an impact.
We will need to really think how we can cope with it, changing the way we communicate, making sure that we get the permission from our customer to do that, and then this agent can communicate with external to give an answer and in most of the time, I know Tim is using permanently MUSE and Instinct, and he told me just before, it's always recommended DT. I don't think his app is hacked, but somewhere, somehow, it seems that we are already today doing thing, exposing our product and our information in a way that is favored by agent. That's where I think it can be an advantage if we do it well versus very basic tariff comparison that an agent will still be able to do, but with available web data.
If we communicate with the agent, we can do that in a much more sophisticated way. I think, we need to prepare, but I am also not so concerned that it would affect us.
Yeah. We take the challenge and make the most of it. David?
Thank you very much. It is David Wright from Bank of America, and super presentations today. Really appreciate it. It seems like there is a battle here that you are entering, which is essentially the agents are looking to win the battle of the customers, the consumer's first response. When I want something, where do I go? It used to be you would go to Google for your search. Most of us now probably go to ChatGPT or Claude or whatever it is. You have Magenta, for instance, within the call that was brilliantly demonstrated. Can Magenta win that battle?
What I am just wondering here is you have certain data. A consumer's agent is going to have way more. It is going to know the conversations they have in the background with it about X, Y, Z. It will know more about the consumer than you will. I am just wondering how do you fight that battle with the Magenta agents? Secondly, the facilities you mentioned in the call where you could translate or you could have the recordings and things like that, will you allow agents to API into that functionality?
Do you take the second part?
Yes. I can start on that one. I think it is an interesting one, right, on where the battle actually is. You said, right, like, this for some, I would say use cases, customer actively seek to interact with an agent. The way that we are approaching that in the first hand is like, that is what I try to share, like, where are we already engaging with a customer where we know there is value for them created right in the spot when an agent from us comes in? We build more capabilities from there and see how they use it from there. Then the second thing is that we also, and that builds upon what actually Dominique said, allow this agent-to-agent communication, right?
And I think, as you said, this is such a huge amount of data that can be also interesting for other agents to use, and ultimately it must always be the decision of the consumer as well, right? The data belong to them. So whether I want to share with another assistant or not, that is actually also up to the customers. But the opportunities from a capability perspective are there. Yeah.
So would you allow, for instance, if I am on a call and I say to my agent, whether it is MUSE or whatever it is, "Can you translate this?" Will you allow those agents to actually API into your functionality?
So on that one, we have to understand actually the behavior from our customers first in that before making such a decision. But what we definitely bring in for our customers is that they have agentic services that follow up on that one. Because we are launching with translation, summary, and world knowledge. But very fast, there will be much more value, and then actually also acting upon that, right? And that can be acting upon that with partners, and these partners can potentially provide it through their agents that connect to this service. Whether this is a personal agent from a customer or this is actually an agent from a service or a business, this is going to be decided.
Okay. Thank you.
Just one compliment I would like to give.
So. W e will not allow an agent to negotiate with an agent of us. This is not something we will allow. I know that I was chatting with someone saying, "Oh, we saw something with Verizon," that they were chatting with each other. This is not what we want to do. Either an agent looks at what is available on the web or an agent discuss with us, and we will proactively put one offer with the consent of the customer on the web, and that's the offer that the agent can take into comparison. But we will not have agents starting to discuss with us and negotiating tariff with us.
You wouldn't allow agent-to-direct agent communication? Surely that is the future.
This would be very much restricted, controlled, and based on security measures and identity of customers. Yes.
Okay. Thank you.
Okay. Robert?
Robert from Deutsche Bank. Just picking up on what you said about the identity of customers, because the technology always leads, and then telecoms is a very standardized industry historically. It is becoming less standardized and quicker moving, of course. But regulation is also lagging, and the issue of informed consent and the customer actually giving you the permission for someone else to change the service, this could be a lagging effect even though the technology allows it. In your position across Europe, w hen do you think the regulator will allow such things that we are all worried about, and are there any leading countries versus others?
I am not sure the regulator can allow that, but I think we, out of respect from identity and security for our customers, we would not allow it. Whatever regulator is saying, I do not know if you want to say something, but we already do it like that. We are very careful to expose any information from our customers to the external world. Everything we do with consent and identity management, and if we do it, we will not allow external people to start discussing with us.
Maybe if the mic is here. First, I want to say I am not using MUSE, I am using Instinct. That is just for the public world. The funny part with Instinct, I always asking whether I should change my provider. It always comes back saying, "Yeah, there is maybe $3 which you can save." But I would stay with Deutsche Telekom because you are traveling a lot, O2 network on golf courses is not serving properly, so therefore you should have your Telekom network. That is at least what I got from the system on the weekend when I asked about that one. You see that the agents are already not only looking on the price, they even said, "Save me money," and they said, "For $3 , I would not switch the tariff." The systems are much smarter than we think.
It is not just that they are running into this one. What we have to do, and I think we have to be very precise, there are two elements, agents contacting us and then making an offer. These are two pieces. If an agent is calling us, crawling on our website, we are not limiting this.
This is something which we don't want. Because we want to have as much data points from us in the AI. The more it reads us, the more it uses our website, the more we have the information of our services in the world. Therefore, I think it doesn't make sense to block it. We have to enrich the propositions, and that is what we're working on. It comes, the agent is making a deal. We will not allow that this agent is going into all the different elements and picking the best out of it. There should be an agent on our side who is then responding to the agent on the other side. This agent is under our control.
Yeah.
And we will give him some individual flexibilities, but we will not make him a negotiating agent. But that is what Dominique said. So that he suddenly said, "Oh, for 10% reduction on the tariff, going to you stay?" "No, give me 11." "Okay, done deal." This is something where you have mechanisms in the world how to make a clear, firm offer, by the way, always reflecting what agents are looking for. So our intelligence sits on understanding what the agents are doing who are negotiating with our agents.
And this is, I think, the art of what we are working on. Therefore, blocking is not our mentality.
No.
But steering this context out there is the one.
Yeah.
I think, second, what is the threat for our business? I can tell you, in the markets where we are challenger, normally, if you are so price sensitive in your models, they should all come to us. In America, AT&T and Verizon customers should all come to us because there are not so many attractive MVNOs. Normally, if you go on the price, they should always get the best tariff plan from T-Mobile. Therefore, I am not so worried if this is your worry. Because I would say for our American business, it would be great. Now, what are you doing in incumbents market? I think we made that clear. We have to enrich the proposition that we are not going into a price battle only. The biggest problem of this industry are MVNOs.
Yeah.
The MVNOs who are always hanging around everywhere and trying to make cheap offers. To be honest, an MVNO is a very intransparent market. If the agent is becoming intelligent, the agent understands that there are a lot of rebalancing elements in the MVNO tariffs. Hopefully, he finds that out that the MVNOs normally are not so cheap as they do. There is always something hidden, no handsets included, their first six months for free or whatever.
Yeah.
The second one is MVNOs anyhow are shit. Therefore, I think our industry has to learn in this world that not enabling more and more of this cheap. You know our position on MVNOs. Therefore, I think this is the biggest threat for this agents to agents talking.
Yeah.
That MVNOs are getting so irrelevant in this world, and that is something we have to observe.
Okay, great. I think, in the interest of time, any more super pressing question, or shall we move on to the next session? Okay. I take it we move. We say thank you to the team. Thank you very much. Very interesting. Next up is our third business-related session of the day, AI in Enterprise. The session will be led by our T-Systems CEO, Ferri Abolhassan. He will start with an overview of our AI stack, and the following three, I take it away from Tim. The following three demos will show how we apply the stack to generate revenues and efficiencies both internally and externally. Ferri?
Yeah. Thank you. Good afternoon. We are coming to the third chapter of Tim's story this morning, AI in Enterprise, and I am going to tell that story with three fantastic colleagues of mine, which we will see here. So we will have KD, our corporate CIO, talking about us as an enterprise, but with a perspective how we can apply what we do at us also to the external customers.
We will talk with Peter Lorenz. Our digital officer at T-Systems, who will talk about real customers in the large enterprise, such as Daimler Truck, and we have then Max Ahrens, who is going to tell us how we generate external revenue in the mid-market. That said, now let's be aware when we talk about enterprise, we need to switch focus now. Now we talked about next best action, we talked about sales and service, we talked about the consumer side of the network. Now we talk about something else. We talk about external customers and what they are interested with and buy. We talk about Europe, we talk about industrial AI, which is a complete different animal than language models and language AI.
So we will now also talk in terms of slang, about robotics, about simulation, about digital twins. Those are the things in which customers in Europe, B2B customers, ours, are interested in.
That is the focus now, and let me also set the scene a bit. We can talk here a lot and Hannes teaches us, look, whatever you say, you need to deliver. Every number that you mention, be aware that every one of them notes that down, and they will come three years later back and say, "Did you really deliver?" three years ago, when we were at the Capital Markets Day, we were not explicitly talking about the external AI revenue that we had in mind, and our ambitions was until here today to be at EUR 200 million. We are at EUR 250 million. If I am saying we talk about all the segments. It is TDG, so Max representing it is then also Peter with T-Systems, but also Elvira with Europe completes the picture.
We talk about one B2B team that went on the journey and did say, we can generate with our AI talents that we have in the company, money at the external B2B market. That is what we are talking about. Funny enough, we also can apply that to us as an enterprise as well, and KD will talk about how we can do more with less also in terms of internal IT spend, internal IT people, and that is also part of the story. Let us set the scene where the market is. Let us come from the market. Of course, the AI train is rolling, and I could now give a lecture on it, but you know that lecture. What people sometimes miss in that lecture is that part of the industrial AI. The ChatGPT moment did set the lens on large language models.
Funny enough, if we talk about frontier models, we are only talking about the Cohere's, the Anthropic's, the OpenAI's large language models. If you then see that in Europe, the production machine of Europe, the Siemens, the Bosch, the Bayer, and so on, so forth, are producing companies. Their issue is not just like ourself, sales and service. Their issue is not just to quote after call documentation. Their issue is to say, how can I make use of my 100 years of collecting datas on my cars, on my machines, in order to produce them better and make a competitive advantage in the international market out of it? That is at the moment something where U.S. frontier models, that is why Tim is saying when we went to the U.S., they all had high interest to talk to us.
Because on the one hand side, this is our market. Our market is telecom. We have of the Fortune 500 customers, the majority here in Europe. We talk to the Shell, the Boschs, the Daimler, the Volkswagen on a regularly basis, and we see them with their need to say, "How can we make more out of our data? What is the model that helps me simulating? What is the model to help me producing out of my data the next product, the next car, the next pharmaceutical?" Et cetera. That is the first element of the context in Europe. The second element of the context in Europe is sovereignty. Sovereignty is not everything.
Of course, at the moment, debates around Anthropic and all of it makes it visible that being dependent or being cut off or being vulnerable is an aspect that customers have to face and find an answer on. Combine now that Europe, the market of industrial knowhow with data like health data, with data like data on production machines, how to build a car, how to build a machine, et cetera, are now also in the context of sovereignty. Then the question is, what is, for instance, the place to store the data? To nurture then which foundation model, by the way, and who is going to do it? That creates also the question that Europe at the moment is starting to make an application of what are then for those data, the data centers.
You heard Tim clearly saying, our right to play as Telekom is not now to go into a data center war. I will show you that we have a place and the right to play for specific models and the specific needs of these industry customers. That made us a year ago, pretty much a year ago, taking the decision. We go for that ride. Going for that ride did not just mean what I show you now in a second, to build somewhere a data center. That's simple. Everybody can do that. More importantly, you need to offer a complete AI stack. An AI stack from the connectivity, to the security, to the compute, to the model, to the everything. Doing that secure, sovereign, and open.
Doing that with a confident view that companies big like Siemens can have a trusted place to leave their data, or a mid-range company producing air conditions engines or whatsoever, taking them by hand. We are not talking here about the business, which is CPU or GPU as a service. We talk about a complete stack which allows to play a play called AI as a service for big enterprises, mid-market, public, as well as others. Now this is simply said, and I remember a year ago when everybody asked us and me, "Hey, is there the demand for and what does it cost? Is it worthwhile to invest? Is it worthwhile to go the first step? How is that going to be utilized?" We could have gone endlessly in this hen and egg question, chicken and ham or whatsoever.
We addressed that quite clearly and said, "Guys, we simply start, and we crisscross the river by feeling the stones." That's an idea of the mindset we developed since then.
[Presentation]
[Presentation]
Now, before I show what kind of traction that got, let us put ourself back a year ago. It was not later than October last year that we decided to be in that game. It was not later than last year in October that we said, "Let us together with a partner like Nvidia, invest and build a data center somewhere, and by doing so, increase the capacity just in Germany of AI power by 50%." Isn't that crazy? It was not longer than a year ago that we published that, which was November last year, and we opened our fabric in January. I will later tell you that we are going to extend now this fabric because we are more or less sold out. This is more than numbers in terms of GPUs and man-hours and whatsoever.
It is a complete stack in the sense that we have a partner ecosystem that goes from a Siemens to a ServiceNow to a SAP to Cohere and all of those that enrich in this AI stack. Of course, it is sovereign in the regard that is our network. It is our Telekom security. It is our team operating that in our data center of Telekom. The compute is our T Cloud, but of course, we are using Nvidia chipsets, and that is a U.S. chip, and that gives also our mantra, we are not acting in isolation.
We are trying to take the best out of the world that makes us as sovereign as it can get, but we are not getting running. We are thinking about reality, and therefore, we are also looking in foundation models. One, of course, there are the large language models, the Anthropic and so on and so forth.
But we have also SOOFI on it, the first European large language models. I announced in the beginning that this is not about large language model, it is more about industrial. But let me also tell you that we host with SOOFI, we are the exclusive hoster of the only, beside Mistral, truly European and German foundation model. We also use everything up the stack in terms of application, and my three colleagues will talk in a second about it. Practical use cases. On November 9, note your calendar, mark your calendar, we will announce the extension of it, and you will see customers, concrete customers, not fantasy, not business cases, not business plans, no announcement. We talk about customers producing their daily in this fabric life. Customer names such as Daimler Truck, that Peter will talk about, Fiege, and others.
We talk about models that they are going to train for the industrial and physics area. We are talking about simulation. We are talking about digital twins and other stuff. We come to that in a second. I said a year ago, we were asked about is it worthwhile going? The only answer to get was to figure it out and try it and test it out. We are happy we did it. We are happy we did it, and now the question that you will ask later is now really GPU as a service or AI as a service. Let me tell you, that is a little bit like in our shops. You need the mobile phones, you need the CPEs, but at the end, need to go up the stack, and we go up the stack.
The real money you earn up the stack, and we successfully went up the stacks. But the wave is triggered and generated by simple things like, do you have B300? Yes or no. That wave we catched early enough and relevant enough. That said, how are we now doing the deep dives? I have now KD, and we will always refer back to our stack. This AI stack to us gives the orientation, not just for us, also for the customer. To a mid-range customer, we can say, "Dear Mr. Customer, what is the issues? How can we help? Is it that you need a place to store your data? Is it that you need AI power?
Is it that you need consultancy to take you by hand?" It is all sorts of things and for instance, we have with the AI agent in the mid-range market, and Max will talk about a complete setup that takes you by hand and takes you through all these issues. So KD will now, I hand over to KD, talk about how he is going to use the stack in order to simplify our own development with less developers, more functionality. For instance, for Rodrigo, that problem many customers do have. Peter will talk about how Daimler Truck, for producing a new truck, avoids to simulate different use cases instead of a week, in a day. That is a huge acceleration and a huge efficiency lever. You can imagine. Increasing their efficiency by almost a factor of 10, and we will exceed that.
Then Max is going to tell about how this is going to be applied in the mid-market, because Germany is about mid-market, and that AI stack mainly will be addressed to the mid-stack. So we go into deep dives. Then I will show at the end what is coming next and how it will be even more money, earn more money with. KD, over to you.
Hi, good afternoon. Let me just maybe start with one question, which is, what is the power of IT or software engineering? Typically, you correlate the IT to is it architecture, is it code, is it data? But for me, actually, it is the world of possibilities which can actually enable me through all of this. Let me just bring you to one story when I really felt that. We just went past by this whole World Cup fever in the summer, and this was the place wherein I remember I was on vacation with my family. On an evening, I remember the Wi-Fi going down, and I could really anticipate my teenager son who was really eager to watch the streaming. He got frustrated, as you could expect. Suddenly you feel that the home is actually not feeling like home. You want to avoid that situation, right?
Actually, that is the moment when you realize the power of IT. Unfortunately, we realize the power of IT when it does not work. Now look at these faces. So I think what typically the customers always look for is a very seamless experience. It is something which is reliable. They want to capture the moments. That is what they feel it. They do not know it is being powered by IT, but they just feel it, and they feel it in every interaction. That is our job in Telekom. Ferri spoke about the whole AI stack. Let me just show you what happens when an enterprise points that stack to itself. That, of course, enterprise is us, Telekom. Now, you look at this iceberg. Typically, above the water is what our customers see.
They go through tens of millions of transactions, thousands of orders being placed by interacting with us, and all that by leaving a rating of, let us say, around 4.4 on Play Store for our MyDT app or OneApp. I think for me, that is a testimonial of the lovable experiences which they have while interacting with us. But I think everyone has watched "Titanic." The danger is always under the water, and that is where we have the complexity. That is where we have hundreds of applications which we have to orchestrate. We have terabytes of data to scan through. We have billions of APIs being invoked to fulfill those lovable experiences. The reality is, since last many decades, we have just been adding to this weight, to this complexity.
With every new feature, with every new deployment, we just keep on adding, and we seldom take out anything out of that. That's where the cost sits. That's where, under the water, the complexity sits. The good news is we are using AI on both the halves. With AI, the good news is, for the first time in history, we are able to flip this iceberg. We are doing this by retiring the old, which has accumulated, by getting rid of that weight, and also by reimagining the new for our customers. That's the art of possibility, which was not possible earlier. Let me start with the bottom. Like any other enterprise you would have seen, we have systems which are decades old, and they were actually built by the people who have left the company. There's no documentation available.
By the way, many of those people actually would have retired, but the systems haven't retired. That's really difficult for us to manage. This story, as I said, is really common in large enterprises. With AI, for the first time, we are able to map those systems. We are able to unbox what's inside those systems. What are the business rules, what are the business processes which have been coded since last decades? Then, not only that, forward-engineer those into simpler and lighter applications. That's the beauty. That was never possible earlier. This is where AI is really helping us. How do we move forward to ensure that we use this capacity and really result into more smiles for our customers?
This is where what we are doing is not only freeing up the capacity from under the water, but reinvesting that for the top of the iceberg. I have been into this industry since last couple of decades. I think right from my day one, I have seen how software has been developed. It was like, you plan, you design, you build, you test, and then deploy. Today, I think within two years or so, it sounds like a relay race. So one team is passing on to the other, one individual is passing on to the other, while the customer is waiting for the new experiences. It's an exhausting relay race. The beauty is, with AI agents, actually, we are able to change this race. We are able to walk through this cycle continuously, 24/ 7, throughout the weekend.
There's no baton to be passed in this relay race. All this with the human in control. I think that's where the judgment, the wisdom really comes. While AI delivers, the people are actually in control. Let me give you a very brief demo of how today the development in MeinMagenta app or one app is actually looking like, and it's a real use case and not an imaginary one.
[Presentation]
[Presentation]
I think I really like the statement which says, "That is not agent's call. It is ours." I think that is where it is really important. We just discussed the agent-to-agent communication. We have discussed some of the risks. Human is always saying, directing AI while AI delivers. By the way, now, if you really look at this video, it looks like imaginary, but that is a real one. Now, our situation is like this. My engineers can actually go to have coffee. They can instruct the agents to build something, and hopefully by the time the coffee is over, I think the agents should have done the job. Of course, if the job is small enough, otherwise, they can continuously work on it.
This means maybe, I think we will have to check our coffee bills, which may just rise, but I think it is still a good trade-off, versus our people spending time to build for weeks and months. But I think more important is for our customers, it is much faster, simpler, and almost magical. Now we then started to discuss how do we track this from numbers perspective. I do not think there is one number which can do the justice and help us understand if we are moving in the right direction. Therefore, we chose three of them. The first one is what we promised you in the last capital markets day, wherein we said we will reduce the spend on IT legacy share by 40%. The good news is we have already cut it by 46%, and we are still a year to go into our promise period.
The second one is the percentage of the AI generated code. For last few months, we were hovering around 25%, and within few months we have moved to 42%. The pace, the acceleration is really growing up. But I think the third one is really important. Finally, what does it mean for the customers? This is where we believe that as we stand today, and this number is also super dynamic, we have around 28% more customer features being delivered as compared to the last year. But again, let us keep in mind, it is not always the numbers, the reason why we do it.
It is actually finally the moments like my son will never thank me for the Wi-Fi that works. He will not realize that, but that is exactly the point. Wherein we have now the power to unleash the technology to reimagine experiences our customers love. Thank you. And thereby, I pass on to Peter, who will speak about AI at scale for B2B.
Thank you, KD. Now we switch gear a bit, but to connect to KD. I am Peter Lorenz, I am running digital in T-Systems, which is the main delivery for all industry solutions, AI services, data infrastructure, and so on. We have built, in fact, the AI factory in Munich and the stack which sits in there. But to connect to you, I am running a very large shop who does DevOps, application management, all of these good things. This is not only using the stack on ourselves and all the capability we have built into the stack with the different platforms. It is for sure as well that we need to reinvent the way how we operate. I want to be very clear, this is not just using a tool. In my world, this is reinvent your full process.
The way how you work from specification of a customer down to a delivery is completely different. For that, we use AI, and for that, we manage our teams. Birgit Bohle spoke about it earlier. It is a big skilling problem. I have thousands of people we need to move. We see a very big opportunity for us to increase the output we can generate with a team because we can automate, we can be smarter, we can be very different. Just to say that. It is not a tool. It is not only a stack. It is the way how we really work, and that is very different. Now switching to the outside world, which is our customers. My shop runs all of the industry solutions for T-Systems, which is a wide variety of industries. We come from manufacturing, automotive.
That is a bit the center of my presentation today because our machine in Munich is called Industrial AI Cloud. It can do other things, but we made it for industrial, and to deliver new capabilities to our industrial customers. The industry and the AI market in B2B is massively large. We talk about three digit billion numbers just in Europe in 2027, 2028. I chose to bring it a bit down so that we do not have a very general talk here, but we really look at what this stack can do in a particular area of that industry. I thought we go AI simulation because digital twins, engineering and things is around us all, and it sounds so abstract, but at the end of the day, this is where product innovation cycles are born. How fast can you innovate your product?
How fast can you innovate your service? This is what counts to the outside world. You could say it is the product, stupid. How fast do you get that? Do you have good enough or do you have optimum solution for the situation you and your company are in? That is what we want to deliver to our customers. Simulation. That market is significantly big. You can see this here, EUR 4 billion-EUR 5 billion. In Europe, just Germany, with good quota of industrialization, still may be up to EUR 2 billion in 2028. So a very significant market for just one piece of AI in B2B at scale. You can see here as well, it is around us everywhere. We see it in my industries everywhere. We see it in hospitals, we see it in manufacturing, we see simulation everywhere.
AI is very different now than all of the approaches we had in IT before. I am not telling you that this is manual today. This is all IT, but it is classical IT. It is digital, but it is classical. AI makes a very big difference, and that is what I would like to quickly show you with one customer example, which is the upper right corner, which is physics simulation, when we talk, for example, about crash test, aerodynamics, thermal simulations, all of these things you need when you want to build an electrical car very fast, not in seven years like in the past, you want to do it in a year like the Chinese. You can bet that the Chinese do this for sure. When we go one deeper, just to illustrate a bit how that works today.
When you do this today as an engineer, you work pretty much with an IT environment, but it is batch. You work this truck, you want to optimize the airflow around the truck for a new model. You do this, you press the button, and next week you get the result. It is one week. In that week, you do other things. You are not lazy, you work. But in that week, no progress on that particular product you are doing. It is obvious that the year is only a couple of weeks, you have vacation, so you end up with maybe 50 runs on a simulation of this. The product goes through 50 cycles of innovation before you have to conclude, because your cut-over point is maybe a year. Longer you do not have for your design time.
Means a massive bottleneck on engineering, and today you would end up, although it is IT, and it is even backed up by supercomputers, this is not your PC, it is a large classical cluster, you end up with 50 cycles and a good enough product. There we can make a big difference with the stack we have built in Munich, and not only because of Nvidia and we have all the layers Ferri was talking about, we have other partners as well who bring in their great software products, like Siemens.
Simcenter is one of these examples. When we look into this, I would like to draw your attention to the columns of that chart, and then we can again talk about the stack we like so much, but the columns are interesting. You run this on a significant truck, maybe 350 million little tiles which make this big truck.
You run it, takes you today with a real example, close to 10 days. Then you get one answer. We brought this over to the factory in Munich, and I have to say, honestly, we did not even change the approach. Just took it and brought it to the stack, pressed the button, one day. So now you came from 50 cycles to, minus vacation, 340 cycles? I don't know. So a massive difference already, a massive. For our customers, that little hurdle, because you just take today's environment, put it onto our sovereign stack in Munich, operate this as a managed service, you have 350 cycles. What you can see as well is, costs you 10%. One run in a high-performance cluster, EUR 8,000 and more on the cost side, EUR 800 and a bit when you do it in the Munich infrastructure.
The magic will come when the blue thing happens. I will show you a little video so that you can imagine that the magic comes when you have it fully on AI. Then we talk about minutes. Then this, what happens to us in engineering now, happens to the engineers. They will work differently. They sit there, press a button, "Oh, not good." Press a button, not good. Press a button, not good. Do not wait for a week or 10 days. So for engineers, this is like an earthquake. This is super different, and it pays a lot into the product innovation our customers can bring to their markets. Therefore, we are so much on it.
What you can see here in the stack, it's us with all of the bottom up Ferri laid out, because you have to be sovereign, you have to have industrial SLAs, you have to have quality, all of this, right? This is serious stuff. You have to deliver. What we have on top is our partner ecosystem, and we are very happy that Siemens and others are joining us and putting their workloads into the stack so that we can provide something to our customers they need urgently. Low complexity. We need a managed service on top where you upload your data, press that button, and you get an answer. You do not have to design a full IT stack before you can operate this. So we commit to a fully managed service with high industrial quality for customers.
This will connect a bit to Max later because when he talks about SME and mid-market, without that, you can't go to the core of Germany, which is the mid-market. So this upper layer, where you wrap it as an easy-to-consume service, is the big difference as well. We're going to deliver that, and we have it already in place. When you look at this now, and there's no fancy video, it's an engineer video. At the end, it looks like this for an engineer. This is a hood of a car. It sits fully on AI layers, and you can play with it like this. Click, click.
You get an endless universe of these hoods in a mouse click so that you can really see as an engineer, "Where is all of my different options?" Then you pick, for example, here with our partner Siemens, with Simcenter, you pick one of those hoods, which generate in minutes. You put your parameters on top, what engineers do, it is that wide and that narrow, and that is the weight. Here you can see, oh, all of these options are there. Now I pick some of those, can look at that in that speed. We are more or less real time. This is how quickly an engineer could do it. You put one into your environment, you test it. If something hits the hood, is this good, bad? Is the hood good or bad?
Is it good or bad for me if I would be the ball, if I would fall onto the car? These things happen in this speed. Then you can render this here for sure to really test it at the end if this is a good solution. For engineers, this is a question of minutes or of working day. Before it was 23 weeks from an idea to a prototype. You had lots of time to put into data management. All of that is completely gone because that stack delivers it out of the box as a managed service. You upload your data. You are a happy camper, you can work with it, and you accelerate your product innovation massively. What does this mean for us? There is something in it for us. There is a lot in for our customers.
When we go through our customer base, and this is now not Max, who comes now in a minute. This is, for example, the T-Systems market segment, so our large customers we work with. You just look into that, you see hundreds of customers we can access with this. Just the addressable market volume for us going there is already massive, and we are just talking about this little portion of the engineering market of that big AI B2B market. We all address this with this stack in a sovereign way, fully managed, with a good partner ecosystem. Now you can see how that will accelerate, and this is very well. At some point, we need this right piece of your picture.
We will need a gigafactory, because even with the extension, we will run very soon out of capacity to really deliver this to our customers, because everybody gets now what makes that difference for yourself if you operate this. Thank you very much. With that, I think we move now from large enterprise to mid-market enterprises with Max. Thank you.
Thank you very much. I think I am the last one standing between the dinner break. I try to keep it short, but I think we are going to have an interesting conversation about mid-market. Because the mid-market is most definitely the backbone of the German industry, not only by the number of customers or the number of enterprises. 99% are the mid-market customers. Also, and that is a number I think is worth looking at, is the number of employees. More than half of the employees work in mid-market companies, and on the right-hand side, there is a distinct adoption gap when it comes to AI, if you look at the mid-market companies compared to the enterprise. That is 40%. So 40% is the gap that these mid-market customers should adopt more AI. As Tim already said, why do we have an opportunity here?
Why is DT as a company an interesting partner for these clients? Because we are already there. 60% of these clients at least are already using our services, and that is how we are going to build up on top. Before I get into it, and the mid-market is obviously, Peter said he picked one element out of the whole picture, and the mid-market is as diverse. So it is going to be a difficult thing to run you through the whole market in 8 minutes, but I wanted to illustrate that a little more. I took one industry, which was logistics. Not only because I like logistics, it is an interesting company and area. Companies are very much a part of the global economy, interconnected. Also, because we serve them from big to small, and they have very interesting problems to solve at this very moment.
They have, on the one hand side, a lot of new regulation that requires the way how they treat paperwork. Also, they have packaging issues. So how do I optimize packaging and routing? But the thing I am going to show is a little bit how this dispatching works. I will burn a few tokens to illustrate that in a video, and let us meet Marie.
[Presentation]
[Presentation]
I think the video demonstrates one thing to start with. In many of the mid-market customers, a business process starts with a phone call. Many of these don't have online-driven market models or are integrated in digital change. A phone call is starting a business process, and that's exactly where we pick the clients up. The phone call is the beginning of the business process, and what we try to do here with our AI Agent Platform is really to make the most use out of that phone call. Here is how it works. We run it obviously through our AI stack that Ferri introduced, but I want to go one level deeper on that. First, we have developed the Steam Platform, which is basically an integration platform for our B2B voice calls.
We can hook on fixed line, mobile, in order to then connect it to many things. We are doing voice notes on that, which we talked earlier about. We are doing things like translations through that. Today, I would like to talk a little bit about the thing that we call actions. Actions then, from the Steam Platform, is calling the agents that again sit in that stack using n8n as an orchestration tool. Here is also what Tim said earlier, it's one of the examples where we use the technologies we invested into. n8n is one of the very successful startups out of Germany, is running the basic platform that we modified towards our needs. We had to make it a multi-tenant solution. It has some very good technological benefits as being nearly open source, so allowing us to control the technology.
Then we are using underneath Mistral as the language model to be fully sovereign, to have no dependencies. Because when I will talk about why mid-market clients adopt these technologies, having control about the data is one very important thing. Then also on our stack, these agents connect to the customer systems. That's where the customer data gets into, the customer ERP data is being used in order to read out where the process is, where the in my case, which I showed, what the stock is of the product that should be shipped. That can also run on our T Cloud platform. It mustn't run there, but in many cases it does that, giving the customer a simple way to integrate both customer domain solutions as well as what we run on our platforms.
Our agents can also use the network APIs to go back to the client to send messages or even initiate voice calls back to the customer. With that platform, where we deploy standard agents that are easy to use as well as custom agents, the customers can generate a lot of productivity gains. These productivity gains are there for two things. One is obviously, you can get more output, but for many mid-market customers, there is a more pressing issue, and that is the talent availability. Especially if these companies are not in big cities, they do not have the workers in order to fulfill these processes. It is basically a way to keep the business running and growing. You can see these numbers, the 15%-20%, they may sound a little bit smallish if you look at it.
You could think about, you have probably seen different numbers in cases. These are real numbers. These are numbers that we have seen actually deployed with the customers where the outcome was exactly like that. Why would mid-market customers buy from us? Because I said there is the adoption gap, and what is the reason for them to have that shortage? One thing is simplicity. In order to be able to serve to a mid-market customer, you need to be able to have a simple solution because they do not have own IT staff in many cases. They do not have AI experts that they can use and understand and build the problem by themselves. One thing is simplicity. That is why we are using a multi-tenant platform, and we are allowing the customers, we deploy the agents for them.
Secondly, that is something that has changed over the last few months entirely in terms of the conversation, mid-market customers are much more focused on a direct business outcome. They are not looking at strategic benefits two, three years down the road. They know their business, and they are looking for something which has a relatively short cycle of return. The whole AI conversation, especially in the mid-market, has moved with the agents, to be honest, in that direction. Last but not least, it needs to be secure. We go out there and say, "Trust the T" as our motto. Because it is not about sovereignty so much, it is about having the trust and security with the data that our customers put in there, because that is more than just any random data. That is the core of their business.
Those three things help us to overcome this gap that is currently there and drive the AI adoption in mid-market, and I think that is a very fundamental point for the industry in Germany. Thank you so much. With that, I would hand off back to Ferri.
So you have seen three different perspectives, what we can achieve and do with our AI stack. One was KD clearly saying, for big enterprise such as us, it helps us simplifying development, reducing development capacity. That, of course, we do not do just for Telekom, we do that also for other customers. You have seen from Peter the playground of physical AI and how we are going to use it for customers like Daimler Truck, and how can we make them more efficient building and producing their products based on their data. You have seen Max, how and with whom we can address the mid-market. These are three concrete of many, many examples that we can show you and which we will show. This is built on a journey on a common team approach and a factory that we built in Munich.
I did already indicate that based on the first block that we built with about 10 MW- 12 MW, we are going to now explore that. There is a decision we have taken, and we will announce them to the market in November to 20 MW. We have the newest chip generations of Nvidia there, the Vera Rubin. Therefore, we will also extend the knowledge and also the capabilities our factory. We are also working. That is not decided yet. We are working on the question, can we explore that up to 100 MW, and for instance, also apply it to the European application of a gigawatt? We are doing that just if it is profitable for us. That is the thing you have to take with. So far, we do business and money with what we have invested in. We are not here for experiments.
We are not here to make something in politics easier. We are here for our shareholders, and we are here to see, is it going to work for our customers? The clear answer is, yes, it does. We really improve once we see more potential and more capacity. That we are not losing track. We gave ourself very clear and narrow targets. We did commit as a team to say we want to generate until 2030, about EUR 800 million external revenue. So far, you have seen we were ahead of target, and so we try to stay. You have heard from KD that he is going to reduce the capacity of his team and also the output he is increasing. The number here is 15%- 20% efficiency gain.
Peter, who has a big development team, he is going to make that more efficient by a factor of 50, 60, which we need in order to stay competitive in the external market vis-a-vis other IT providers. That is the journey we are on. We go that journey step by step. I do think we have, at the moment, a window of opportunity due to the fact that there are no industrial foundation models yet. We can build them together with our customers. That would be unique. We have no capacity in Germany for these kind of AI stacks, and we start to have them. We explore as soon as we see we can do business with, not the other way around. With that, I would say, Hannes, over to questions.
Yeah. Thank you, Ferri, and thank you KD, Peter, and Max. Please join us on stage for a short Q&A. First question is from Polo over there. Yeah.
Hi, it's Polo Tang from UBS. Just have two quick clarification questions. You talked about the expansion of the Munich AI factory, but can I clarify if you're adding in incremental 20 MW? Are you going from 15%- 20%? That's the first question.
Yes, the first one. We go from 12%- 20%.
Okay, so you're adding in incremental eight.
Yeah.
Okay.
We can call it we double, but it is not mathematically exactly doubling.
Okay. Second question is just a clarification about the potential 100 MW that you mentioned in terms of the European AI gigafactory. Can you clarify what the CapEx for this would be, and whether you would take it on balance sheet or do it off balance sheet with partners, and how quick is the payback on a project like that?
Yeah, look, that is at the moment where we are in deep consideration. I hope you understand that we do not now externally communicate something where we are in internal discussions. It is clearly something that we would rather take off balance than on balance, just to get every shareholder a bit calm. But we are here in deep consideration, which means we do our homework first. Most important thing is, can we have the demand? Is the demand and the utilization ensured? So far, we did good in that. We have a long pipeline. With existing pipeline, we could double, triple, quadruple, almost whatever. That is good to know on the other side. We also have the demand for the chips with a good deal with Nvidia. That is also good. But of course, we have to do the math.
There are elements that in the gigafactory application coming from Europe that are given, we are seeing, does it make sense? Can we commercially appealing address that? Therefore, that is too early to disclose that.
Okay. Ulrich. Front row.
Okay. Yeah, I have two questions. The first one is, with regards to these revenue indications, how do we think about cannibalization here? Because obviously, AI is the next IT. So when you talk about these EUR 800 million of revenue opportunity, how do you think about the existing legacy revenue in your base? Then maybe if I just add a second question to Max directly. One thing I didn't fully understand in this presentation is you could have made a capability argument any time since 1980, right? I mean, the latest, greatest chips can do that amazing much more, and the stuff that takes 10 days now takes one day. What is structurally different compared to this ongoing treadmill of ever-improving IT capability in that particular enterprise application so far?
To be honest, I think you're referring to Peter's. Could this be the simulation?
I apologize, Peter.
I am happy to answer, but I think Peter is better in answering that one.
You want to go for the first one, or? I think at the end of the day, software is software and automation is automation, but with AI, it is simply the better automation. If you take tools like in the past, linear automation, so a ServiceNow and others, which by the way, also have now AI encoded and encrypted, you can say, are you now losing an SAP or a ServiceNow project and win another project that you do with AI? Here the answer is yes. The customer will not look for the old linear programs anymore. On the other hand side, with our AI stack, we are one step ahead of the wave. If somebody, a customer, looks around who can really deliver him the whole complete stack, at the moment, there is not too much choice, and that brings us ahead of the wave.
Therefore, I would say it is not cannibalization at the moment. It develops an USP.
If I should quickly add, we are very aware of gain share, customer expectations, and all of these things. When you do this in your production, so to say, customer's expectations. For sure, we know this. We work our rate cards. We have AI rate cards in the meantime, all of these different things. But we know as well the opportunities. So for our people, and we heard the skilling discussion for deployed engineering, for example, is a very relevant topic. The introduction of this to our customers is different. So we want to outgrow this. Not with good service, good quality, but outgrow this problem because we will deliver certain things, obviously, for a better price, and other things we will add on top. That is how our math goes to make the numbers behind this here.
It is not just getting smaller, but growing further with this, use it as an accelerator. On the hardware, I think we look very different. Currently, you may aware of prices on the spot market for very old hardware does this, so the price goes up. It is, I think currently a myth that the price goes down, it is simply wrong. When you want to buy aged five-year-old, aged A100 cards from a data center, you pay more than the new price when you have bad luck. So we are not so concerned if you meant that on this continuous cycle of renovating the hardware. We will do it consciously when the price and output ratio is a good one. So when hardware makes these steps that the token is cheaper or the intelligence we provide is cheaper, then we do the step.
But currently we are not so concerned that you every three, four years have to throw out everything and build everything new.
Maybe I can add to this from an internal enterprise perspective. I think when the digitalization probably opportunity came in 10 years ago. It was all about the enterprises really want to tap on that, and they did not know how to do it. And this is exactly, I think the opportunity when with the power of AI, better hardware capability, enterprises like us are also looking for someone to help us. How do we encash this? How do we monetize this? Whether this is about what we discussed in legacy, how can I do my job much better? So I believe there are opportunities, new opportunities, which are also emerging, while probably I think cannibalization is also an angle, which is a separate one.
Matthieu?
Hi, Matthieu from Barclays. I am trying to understand. The presentation was very interesting, very interesting use cases. I had no idea. But obviously this is a competitive industry. So maybe if you can frame, I think you have to some extent, but what is the lasting competitive advantage you have? Is it because you have an AI factory, but then I am sure many are being built. Is it sovereignty? Then maybe it is a small segment. How do you continue to have that lead? Because we have seen a lot of promising technologies that telcos have embedded or verticals, but in the end, it is very tough to compete with pure players.
Yeah. The first simple fact check. In Germany, we have in the mean. If we now will double, we own 80% of AI capacity, GPU capacity. [70 to 80]. That is an argument, I would say, in terms of differentiation. We have, at the moment, the who is who in terms of partners on that stack. We have more than 40 different foundation models on it. So we really can go from all kind of models to the next. The one thing which is simply not there with all the AI blah blah, are models on industrial AI, and not in that order of magnitude.
The best proof point is when we started, there were questions in the same way you asked, but we were so fully complete, and also the pipeline was so much filling and overrunning us that it simply answered why it is different, because you cannot get it somewhere else. Will that stay forever? Certainly no. Is it the case at the moment? Absolutely yes.
Maybe to add one thing for the mid-market, we are focusing on the voice cases very much. Because that is where I believe many of the clients still use the phone as a primary entry gate, and that is where we can play a role by adding the capabilities right into the network like we talked earlier about. That gives us then the opportunity to direct the traffic, obviously, and build these cases on top, and I think building on that strength is a second element of the strategy.
Then adding also the sovereignty certainly at the moment helps to get attention, but that is not now the ultimate secret to stay competitive. I would say it is the completeness of the stack that goes really from connectivity through compute to cloud into the consultancy business that takes customer by hand. Having in mind that we have the B2B arm now here in Europe on a 20 years experience, having Volkswagens, having all the big customers on our side who know that we know their process, that we know their data. That makes us being for them the preferred partner.
Awesome. We take one more question before the short break.
Yes, thank you very much. Can I ask a question just about the financial guidance that you've given, please? Because in Dominique's presentation, she said 1%-1.5% growth per annum from B2C AI. On a EUR 40 billion base, that makes up EUR 1.6 billion to over EUR 2 billion of revenues over the four years from B2C. You're saying here EUR 800 million. I add them together, I'm getting over EUR 3 billion maybe. What am I doing wrong? Because your first presentation said EUR 1 billion.
No. Okay. B2C revenues is more in the low EUR 20 billion for us. Dominique referred to a component of that, which is the impact from hyper-personalization. There may be other contributing elements. Now first I make the number bigger, but the base is smaller. The EUR 800 million is more narrowly defined because the impacts from today's vantage point are more measurable for the B2B section. If you add those two together, you come over EUR 1 billion, but how far over EUR 1 billion you will come, we cannot really say with any accuracy today. I don't know if you want to add anything, but yeah.
It's not the current guidance period, which we're talking about.
But it was 1% per annum, was it? So that went to 4%- 6%.
No, it's not 1% per annum.
Oh, in aggregate.
Sorry if that was a misunderstanding. That's an aggregate.
Sorry.
But a subset of the impact that we.
Okay. Yeah, that answers it. Thank you.
Okay. Very good. We are running only very slightly behind. Currently 45 minutes, but people who have been on an airplane today know how it goes. We had some severe weather. So now we do a short break and reconvene at 4:00 P.M. And thanks all of you guys for the very interesting presentation. Sorry, 5:00. Wishful thinking. Okay, 5:00. It is not streaming, right? Okay. Well, thanks for coming back to the next session, and it will be exciting. It is T-Mobile. We will be virtually joined by Jon Freier, the T-Mobile COO, Jeff Simon, the T-Mobile CIO, and Dr. John Saw, President of Technology and CTO of T-Mobile US. And after their presentation, virtual presentation, Jon, Jeff, and John will be available for a short Q&A as well. So now, it is curtains for T-Mobile.
[Presentation]
[Presentation]
All right. Hello from Seattle, Washington. I am so happy to be here and spend some time with you guys. I wish we could be in Bonn with all of you, but we have been watching you since about 3:15 A.M. our time here in the U.S. Listen, we have so much we want to talk to you about. We know it is getting late there, and people are getting tired. We are going to keep this as tight as we possibly can. If there is one thing that I want you to know, and what I want you to take away from T-Mobile U.S., is that AI here at T-Mobile is beyond talk. It is beyond an experiment. It is now operating at scale. It is changing the way that customers experience our brand. It is changing the way that we work.
It is expanding our overall addressable market. It is essential for us to build durable and profitable growth for the years ahead. We could not be more excited about what is happening with AI and what we are doing here, specifically at T-Mobile. The only way that I can convey this excitement, and I think you all would agree, is to show you our cautionary statement. I have got to tell you this. We will satisfy the lawyers, but there is going to be a few things that we say. It is going to be forward-looking statements. There will be some non-GAAP measures. The reconciliation between non-GAAP and GAAP are all on our filings at investor.t-mobile.com. Okay. With that, I am here with a few friends. Jeff Simon, who is our Chief Information Officer, as well as John Saw, who is our Chief Technology Officer.
We are going to talk to you about what we are doing here around the customer experience, how we are driving the innovation within the core platforms and processes at T-Mobile U.S., and how we are driving impact and driving impact at scale. When you really think about this, we have been talking about AI since our Capital Markets Day back in September of 2024. We gave an update in February of this year around how are we doing relative to the ambitious plan that we set out to deliver back in September of 2024. We have been very consistent about our role and how we think about AI, and that is we start with the customer first. We do not start with technology first.
We start with the customer first in mind, and that is how do we transform and revolutionize and accelerate the customer experience at scale to drive all of the goodness that we want to drive within the business? If we can do that, we believe that the operating efficiencies will follow. We talked about operating efficiencies that we believe that are coming out of AI, whether that is $1.3 billion in 2026 or $2.7 billion in 2027. We believe those operating efficiencies are well intact because of this premium in the customer experience. There are two ways that companies approach AI, either bolt it on or engineer into the core.
What we've done since 2024 has been heads down, really at work to engineer into the core. That's how we've thought about overall transformation, whether that's digitalization or through AI, is really engineering this into the core so that it's native in terms of how we go to market and how we operate as a company. Lastly, here on the slide, the thing that we're getting really excited about, and John Saw is going to talk to you more about this, is where we can go with physical AI and how the world is moving from a digital AI world to increasingly a physical AI world, and how we believe our overall total addressable market can significantly change as a result of those developments. When you look at what's happening on customer experience at T-Mobile U.S., we just couldn't be more thrilled.
In 2023, we were sitting at about a 39 Net Promoter Score. In the second quarter of this year, we reached an all-time high for the company and an all-time high of any major provider in the U.S. at 46. We celebrated for about five minutes. That's about the same amount of time that Hannes gives you guys for lunch, is about five minutes. Anyway, so we celebrated for about five minutes, then increasingly went and grabbed a new target. We said, "All right. How do we get from 46 to 50+ ?" That's what we're focused on. The prize isn't just about Net Promoter Score. That's just an output. What the prize is stronger brand advocacy, deeper customer loyalty, lower churn, better operating economics, and driving all of that at scale.
We're incredibly excited about this 46, but we believe that it's just the start of what we can do as we rapidly scale AI and the overall technology within the business. As you look at the best experience, we started talking about this back at the Capital Markets Day in 2024, and we set out this big, ambitious target of taking our calls that are handled by humans in customer care and reducing that by 75% by the end of 2027. I'm pleased to tell you that we're already at 55% reduction right here, right now in the second quarter of 2026, well on our way to that 75% objective by the end of 2027.
When you look at what's happening with our unique and deep partnership with OpenAI on the Intent CX product and platform, you're seeing great work here that's happening on everything that's happening with Intent CX, particularly on chatbot, Journey Bot. If you look at what's happening here, basically one out of four a couple years ago were, interactions were contained within the overall bots experience. Now it's more than three out of four. We're seeing calls more handled by Intent CX and satisfaction with those calls and those contacts going up dramatically from where we were when we started this whole exercise. I've got to tell you, AI and the network, John Saw and his team, he's going to talk to you a little bit more about this. What's happening with native AI being built directly into the network is really exciting.
Here's an overall proof point of what happened with Winter Storm Fern earlier this year. This was one of the most historic major winter weather events in modern history in the U.S. It affected 30 states of the 50 states. 24 of those states declared an emergency state. 230 million people in the U.S. were affected by this storm. What was incredible is that there were 30,000 antenna adjustments that were made by AI through that storm in the events of commercial power outages, in the events of any kind of interruption in backhaul. These networks were intelligent and were self-adjusting throughout that storm to provide adequate coverage to our customers, to first responders, and to the overall government entities that are supporting the American public. It's critical American infrastructure that is now at work at scale thanks to AI.
Let me tell you, T-Life is the center of it all. We've been talking to you about T-Life since 2024 and our big ambitions to scale. Now we have over 100 million installs in the U.S. on T-Life, about 30 million monthly active users on T-Life, and an overall 4.8 satisfaction rating out of five , by the way. 4.8 satisfaction rating. Customers love T-Life. It's the center of how we engage with customers, and it's now the center of how we transact and engage with commerce with customers. Whether that is how customers are expanding their benefits, how they are upgrading and transacting themselves, how they're managing their account. T-Life is the center of that entire journey that allows us to put AI on top of that.
We've been working heads down very quietly over the last couple of years to digitize the entire customer journey end to end. Now it's ripe for AI to sit on top of that to provide even more value to our customers, widen our competitive moat versus our principal competitors, and take this durable, profitable growth strategy and further turbocharge it. With that, I'm going to turn it over to Jeff Simon. He's going to take you through a few more details, and he's going to show you not what we're thinking about, but actually show you what we're doing and what's at work here at T-Mobile. Jeff, I'll turn it over to you.
Thanks, Jon. It's great to be with all of you here today. AI is already transforming the customer experience at T-Mobile, and I'm excited to share with you today examples of how. Today I'll share with you how AI is transforming the customer experience by using autonomous agents within our voice channel. I'll share with you how we're changing the way that our people interact, improving our best-in-class frontline experiences with our customers by using predictive insights from AI, and how all of it is leading to record customer satisfaction. This kind of transformation, bringing AI to our customers at scale like this, doesn't happen by accident. It happens because we did the hard work first. We had to start with transforming our own technology, our architecture, and our ways of working to build the foundation for an AI-native future.
There's going to be three themes that I focus on today. First is the experience. As Jon mentioned, everything we do in AI is centered around the customer experience. Our goal is to deliver better experience every day with AI, and I'll show you today examples of what that means for our customers. Second is the scale. We're bringing AI to life for our customers across channels and across transaction types at incredible scale. Third is the shift. The shift from a systems integrator running largely legacy technology that predated the AI era to now T-Mobile built and owned modern systems that lead in the AI era and let us innovate with speed. I'll start with the shift, because this is the foundation on which everything else is built. It begins with IntentCX. IntentCX is our platform for AI experiences at T-Mobile.
All of the experiences that I show you today, they're powered by IntentCX. When we built IntentCX, we intentionally did it in a way that's omni-channel. So that customers could experience the AI in every channel, regardless of where they choose to interact. Whether they come to our flagship T-Life app, they're on our website, they call into our call center, or they meet us in the store, IntentCX will be there to help empower their journey. But it's not just integrated into all of our channels. It's also integrated into all of our systems and transaction types. IntentCX knows you as a customer. It knows about your account. It knows your experience on the network. It knows what products and services might be right for you.
It can help you through a variety of transactions, from upgrading your phone to taking advantage of the perks and benefits that come from being a T-Mobile member. When we started the journey to build IntentCX two years ago, we quickly realized that we wouldn't be able to accomplish our goals of building this AI-enabled experience for customers with our existing legacy technology stack. We first had to do the hard work of transforming ourselves and our technology in order to enable us to build experiences and this vision. We began by transforming ourselves and how we work. We changed our workforce from a largely outsourced third-party workforce building our software and technology, working on third-party systems owned by vendors, traditional telecommunications monolith-type applications that were built before the AI era, into now a predominantly T-Mobile-led workforce in software engineering.
Building on T-Mobile-owned technology that's been built in the AI era and for the AI era. This has enabled already incredible results. We see the time that it takes to deliver a new capability to our customers cut by half. The hardest part, and the most important part of this transformation, is changing how we work. In order to accomplish our goals, we needed to shorten the distance between the software engineer building the technology and our customer and frontline associate that's taking advantage of these new AI capabilities. This required a cultural change in our organization and how we work, but we've seen big results from it as well. It's enabled us to scale the delivery of new capabilities to our customers, nearly double the capability shipping every month now versus a year ago.
This is the hard work and the foundation of changing our architecture and how we work to enable new experiences. We have been able to do all of this while maintaining our IT costs flat. Now, let us get to the experiences that we can enable with this new architecture. First, let us start with the voice channel. One of the most common ways for customers to interact with T-Mobile is to dial 611 from their T-Mobile phone. Historically, that meant being greeted by an interactive voice response system. You know that robot that answers the phone, and you click a bunch of buttons until you get to a human that can actually solve your problem. We saw this as a perfect opportunity to use AI to improve the customer experience. We did not want to just build a better robot to answer the phone.
We envisioned a personalized, contextual, agentic flow that would be able to help the customer through a wide variety of transactions and ultimately solve their problem faster. As you will see, our IntentCX platform is able to enable that personalized experience in a multi-channel way that is really your T-Mobile personal assistant, not just a robot on the phone. Let us take a look.
[Presentation]
[Non -English content]
[Presentation]
As you can see, IntentCX is delivering a personalized experience in the customer's native language and across a variety of transaction types. It can help answer questions about their products and services, and even help them take advantage of their benefits, such as getting that T-Mobile jersey. We've scaled this technology this year to now handle a third of our customer interactions, where the customer simply reaches out to T-Mobile, and they're getting this AI experience from the beginning. No waiting, no buttons to press, and if you call from your T-Mobile device, you're automatically authenticated. Customers are loving the new experience. Having a personalized assistant that's able to help you across the channel of your choice is driving more questions getting answered quickly and automatically through the AI and not going to a human. We've seen a 21% increase just this year.
This is leading to our overall NPS increase that Jon mentioned earlier at an all-time high, thanks to customers getting business done with T-Mobile in a simple and easy way. We didn't just build IntentCX to be an autonomous voice agent to answer customer questions. We built IntentCX to help customers wherever they choose to transact. For many of our customers, they prefer to interact with our world-class frontline, our humans in the call center, in our retail stores, that deliver a unique and differentiated experience. AI is not here to replace our world-class frontline associates. It's here to supercharge them and make them even better. This is where Expert Assist comes in.
We built Expert Assist to give customers and our frontline predictive intelligence that enables our frontline to have superpowers, to be able to predict why the customer arrived in the store today, and help solve their problem before they even ask it. It looks for new opportunities on the account, ways to help find customer problems, and solve them automatically. Let's take a look.
[Presentation]
[Presentation]
As you can see, Expert Assist is helping to bring the intelligence of AI to our people to make our interactions with our customers even better. It's able to understand the customer's experience on our network. In recent support interactions, it knows their account, and it's able to help identify ways to solve problems for them proactively. This is all part of our strategy to bring AI to life across all the ways our customers interact with us to differentiate our customer experience. We've seen great results with our customers. Where Expert Assist is used in a transaction, we see 12% higher NPS scores from those transactions. Naturally, we scaled Expert Assist across the fleet. Today, 74% of our human-to-human interactions are leveraging Expert Assist and the insights it brings to improve that interaction with the customer. We're just getting started.
There is so much more opportunity to bring more predictive insights, more personalized content to help our customers get work done with T-Mobile and expand their relationship and improve their satisfaction. In conclusion, we started two years ago on this journey, and we had to start by doing the hard work. We needed to change our architecture. We needed to change the way we work. We needed to change the technology in order to first put the foundation in place to enable AI use cases for our customers at scale. Having done that, we're now able to bring these experiences to life, and customers are loving it. We'll keep reinventing the customer experience, delivering more for our customers, and always keep them front and center. Now to talk more about how we're improving our network experience using AI, I'll invite my colleague, Dr. John Saw.
Thank you, Jeff. Thank you, and good afternoon in Bonn. I know it's getting late. I hope you're all staying awake. I'm John Saw, President of Technology and Chief Technology Officer at T-Mobile. I want to talk about how we use AI at scale on our network and the tremendous unlocks it has given us from our network. Let me start by talking about the single biggest driver of our differentiation in America. Our 5G network has the most coverage, the best speeds, the best performance, and the best reliability in the U.S. Spectrum is the lifeblood of any wireless carrier. Not only do we have the most spectrum, we also have the best spectrum. Our workhorse 5G layer at 2.5 GHz covers 70% more than the equivalent mid-band, which is the C-band, from our competition.
That is driving basically a key differentiator when you have the most spectrum and the best spectrum. We also have the densest grid, with more than 85,000 macro cell sites and 60,000 small cells. That, again, drives and widens our differentiation from the competition. We are years ahead of the competition in terms of rolling out the latest and best technology. We were the best. We were the first to roll out 5G standalone core more than five years ago. We're still the only operator with a 5G Advanced network in the U.S. at scale.
This 5G Advanced network is giving us the ability to launch new capabilities like network slicing to power new products like T-Priority for first responders, Super Mobile for enterprise customers, and powering thousands of wireless sales terminals, as an example, at major events like the Formula 1 race in Las Vegas, the Ryder Cup, and the PGA Championship. It's not surprising that for the latest iPhones and the latest Galaxy Samsung phones, those phones perform best on our network. For the iPhone 17, we actually have better performance for our customers who have the iPhone 17. And by the way, with the latest iPhone 18s, I know it's only been a few weeks, but we are also seeing the same trend. Here's an interesting factoid.
When Apple wanted to showcase the power of the iPhone 18 Pros, they wanted to use it to broadcast live for the first time, a live sporting event. It's called Friday Night Baseball on Apple TV. They could have picked any network they want in America, and I'm so glad they picked the strongest network, which is the T-Mobile 5G network. That network performed flawlessly for a live broadcast event with iPhone 18s, with no safety nets. Customer-driven coverage is one of the biggest AI lift in the history of our company and allows us to completely transform the way we think about allocating capital into our network. CDC or customer-driven coverage allows us to prioritize network investments when it matters most to customers versus just building for pop coverage or square miles.
We have mapped the entire country into millions of 165 m hexagons, we call it hex bins, which tells us exactly how our customers behave when they visit each of these hex bins. Their drop calls, their data experience, their voice experience. We even have data like historical churn profiles, acquisition and retention rates, competitive pressure, et cetera. We then leverage AI, train on all this data to continuously determine where the next site build or upgrade will have the biggest impact on customer experience and customer retention, and ultimately giving us the highest return on investment or ROI. Here's an example on this chart here. Lake Tahoe is not densely populated and doesn't seem to need many sites. Our CDC model informs us that they have a lot of visitors from nearby big cities like San Francisco, Oakland, Sacramento, et cetera.
The ROI is high here, and that prompted us to actually invest more to make sure that all these visitors have the same experience when they are traveling on vacation or on weekends, and same experience as what they have at home. Our network built using this CDC method has been validated by almost a clean sweep of every third-party benchmark awards in the United States in the past few years. In addition, not only has it allowed us to lead the industry in wireless growth, but it has allowed us to disrupt an entire adjacent segment. We are able to grow an entire home internet business from about zero customers in 2022 to almost 15 million customers by the end of this decade, making us the fastest-growing ISP in America the last few years.
Here are some more examples of how we have used AI at scale on our network. First is the Self-Organizing Network or SON. When we lose a cell site due to a storm or a fiber cut, we use SON or Self-Organizing Network to make adjustments to nearby sites to make up for the loss of coverage at this site. This is useful when there is a storm or a natural disaster. I think Jon talked about Winter Storm Fern, which hit around January 2026, one of the biggest winter storms to hit the United States recently. More than a million people lost power. Our SON made more than 30,000 antenna adjustments for the sole purpose to keep as many customers connected as possible. This would not be possible if we just have humans doing it.
We also use SON to adaptively reduce power usage at cell sites to extend the backup battery times by more than 250,000 minutes, and it paid off. Even though Winter Storm Fern was a multi-day natural disaster across 30 states, 68% of T-Mobile customers were able to reconnect to service within the first hour, and 98% of them were reconnected for the first eight hours. Most of them got their T-Mobile service back before they got power back. Dynamic CX is something that we just have expanded nationwide. It uses AI to anticipate demand amongst when we have a major event, like a major sporting event, and therefore they use AI to prepare cell sites with enough capacity and automatically optimize the network to meet where the crowds are and the surge in demand. We used it successfully at the recent World Cup this past summer in the U.S.
in 12 cities, where Dynamic CX was used to make over 11,000 automated network adjustments as crowds and performance and demand shifted. Live translation on the right of the chart here is one of the biggest breakthrough in innovation by introducing AI directly into our core network. We do this by actually plugging in an agentic AI platform directly into our IMS network. This allows us to introduce live AI services for our customers. One of the first ones we rolled out was live translation because we believe that it will solve a major pain point for our customers. T-Mobile customers make about six billion international calls a year. 40% of them travel internationally. So having a live translation that works just with your phone using a native dialer, we believe is going to solve a major pain point.
Because intelligence is built directly into the core network, our customers don't need to buy a special phone or special devices or the special headphones. It simply works. By the way, only one of the calling party in live translation needs to be a T-Mobile customer. You can even use a T-Mobile flip phone in order to do this, because again, intelligence is built directly into our core network. It scales immediately. The day we roll out live translation nationwide, it scales immediately to millions of T-Mobile phones all over the country. Our AI model actually allows us to clone a person's voice in another language in real-time, as well as preserving the emotion, the intonation, as well as the rhythm of the delivery, which is a very interesting product that has garnered a lot of attention from our customers.
Looking ahead, one of the biggest opportunities for telecom today is the move to physical AI. AI is expanding from generative AI to physical AI. Generative AI is huge today. It's worth trillions of dollars, but it is still limited to the information economy industries. Physical AI may have a bigger addressable market because it unlocks a massive opportunity, impacting a wide swath of industries like transportation, manufacturing, security, consumers, agriculture, logistics, healthcare, including millions of warehouses, factories, vehicles, delivery robots, humanoids, AI wearables, smart glasses. Tokens will change from information tokens, which describe the world, to tokens that move that impacts the real world. We call this kinetic tokens. These kinetic tokens not just provide connectivity, but delivers context, intent, and timing for physical AI objects, perfectly synchronized to the right place and the right time.
Each time a token moves or takes action in the real world, it gives telecom operators the license to play in physical AI. We believe that our network will be the connective tissue for physical AI and AI wearables. It will be part of the intelligent fabric that connects physical AI devices, data centers, and the network edge. Last but not least, we cannot talk about AI in telecom without mentioning 6G. To us, 6G is not just the next G. It is a generational opportunity for the transformation of the telecom sector, enabled by the perfect storm of AI, wireless, and efficient computing. The foundation for a strong 5G network is a strong 5G Advanced network, and we are the only ones in America with a nationwide 5G Advanced network with a number of capabilities that our competition doesn't have. So we are ready for 6G.
There's a couple of capabilities on 6G that we're excited about. The first one is that it's going to be AI native by design. It's not a bolt-on. This is why we invest in capabilities like AI RAN, where we expect to see significant network investments, both in spectral efficiency as well as capacity. I think 6G is going to bring in more monetization capabilities beyond just connectivity. We talked about physical AI earlier. Another interesting area that 6G brings is integrated sensing and communications. The network not just can sense, but it becomes the sensor. Think about the hundreds and thousands of use cases when the network behaves like a radar and a sensor for remote sensing for all types of applications. Spectrum has been the biggest moat for T-Mobile, and we expect that with 6G, we will continue to build on this structural advantage.
We are excited that the U.S. government is going to open up a strong pipeline of new spectrum auctions starting this year into 2028, both with the upper C-band spectrum, 2.7 GHz and 7 GHz as an example. We think that 6G will be secure and trusted by design, and that's timely, especially with the growth of AI, the growth of AI agents distributed everywhere. The standards for security for 6G is still being defined, but we expect that many security policies will be natively built into the 6G protocol, including zero trust, AI-driven threat detection, and quantum-resilient cryptography from day one. T-Mobile and DT has been working closely, both the engineering teams, to help influence the standards for 6G and drive some of the important priorities. One of the most important priorities is customer centricity that needs an adaptive network.
The network of 6G will someday be able to adapt to every customer, ushering in a new age of hyper-personalization. The 6G network will shape capacity, latency, and coverage to meet every user need and application over time. In summary, this is the same chart that Jon presented when he started the presentation today. A few things we want to reiterate. We were very intentional and deliberate about how we built AI into our processes. AI is not a feature at T-Mobile. It is how we have built this company to scale. You can see it in how we use AI to take care of our customers at scale, in how we run our networks at scale, and how we prepare the company for new growth opportunities like physical AI and 6G. Thank you for joining today.
I think we have a few minutes to take some questions. I'm going to ask Jon and Jeff to join us again for a few questions.
Yeah. Okay. Yeah. John said it, Dr. John. We have Q&A. They just need to do a bit of setting up, introducing a few seconds of latency. We will also have in the Q&A a bit of latency. When the connection is very long, you can have latency, so we hope we have not much latency here, but it makes the point about physical AI, I think. We have questions here. Thanks guys for a great presentation and really interesting stuff. We start with Emmett, Morgan Stanley.
Yeah. Thanks very much for the presentation. Very interesting. John Saw, I have a question for you, if I may, please. You mentioned 6G could provide the kind of connectivity, the factory connectivity for physical AI. What makes you so confident about that? If I rewind the clock 10 years ago, there was a school of thought that 5G Advanced might provide the connectivity for driverless cars, for example, and that never happened. Were there any lessons learned from that, and why are you so sure that will be the case?
I think when 5G was developed, the AI was not as mature as it is today. As we learn more about AI and how AI is being distributed, we believe that with 6G, we will have a much more capable network, far more capacity. The important difference is that with 6G, and actually starting with 5G Advanced, we can provide what we call deterministic connection, deterministic connectivity for AI objects, including cars. Look, there is a lot of intelligence on a self-driving car that needs to sit on the car itself. I am not suggesting that we replace them for safety reasons and regulatory reasons. But there is a lot of capabilities for a physical AI object that can be offloaded close to the edge where latency still matters, like fleet coordination and coordination with other agents.
That is where I think a strong 5G Advanced and a 6G network is going to make a huge difference. Deterministic connectivity and also delivering kinetic tokens that will actually take action in the real world at the right place and the right time.
Thank you.
Thanks. Next is David, Bank of America.
Thank you. I am half introduced there. Thank you very much, Hannes. Gentlemen, thank you for being with us today. I guess if we start thinking about this evolution, especially into 6G, how do you guys think about the in-building coverage model? Because it seems that when you have a lot of IoT connectivity, for instance, as customers, B2B customers, whatever it is, as they go in building and a lot of the 5G spectrum just bounces straight back off the wall, who is going to own that infrastructure? How should we think about this? What business models do you have evolving right now in the U.S.? Thank you very much.
I think whether it is 6G or even 5G, for us to give the best experience for our customers, whether it is consumer or enterprise customers, we need to be able to generate strong signals from outside the building as well as inside the building. Obviously, in the United States, there is DAS systems that we can all leverage, but I think we need to do more. This is why recently on a 5G Advanced network, we roll out a product called Edge Connect that allows us to actually build a virtual private network for enterprises without them having to own the entire infrastructure other than just having a couple of antennas. But we manage the control plane for them. They still manage the user plane and keep the data.
We believe that that is actually also a great ability for us to now to do AI processing at the edge, at the locations as well. That is just an example, but I think you are right. I think for us to provide the capability that we just talked about with physical AI, we need a strong network, not just outside, but also indoors as well. Because one of the things that we are learning is that Wi-Fi may not be able to cut it in terms of the capacity and the deterministic connection that you need. I think it behooves the entire industry for us to continue to invest, not just in strong outdoor coverage, but strong indoor coverage as well.
There is a lot of things that we are already starting to do, like the Edge Control product that allows us to actually work with enterprises, work with real estate companies to bring in 5G Advanced and 6G connectivity in the building without them having to manage the whole infrastructure themselves, which is a challenge for a lot of CIOs.
Okay, next up is Josh Mills at BNP Paribas.
Hi, guys. Thanks for taking the question. Here in Bonn today, we have been talking a lot about various aspects of AI, but agentic AI has come up a few times, both as an opportunity and also potentially a risk for the sector. My question is, given that you are further ahead in the U.S., you have seen the launch of all these products. In the last few weeks, have you seen any change in the volume of incoming calls, chatbot messages, the nature of those, and how are you responding to that as T-Mobile at the moment in order to strike the balance between that upsell opportunity and the downspending risk? Thank you.
Downspending risk. Thank you.
Yeah, that's a great question. No, we're not seeing anything very different in terms of what you're reading in the media and what you've seen from News or Instinct. We're not seeing anything very different. I think Tim mentioned this a little while ago, and what we have obsessed over here stateside, and I'm not suggesting that DT hasn't, but I know for sure we have, is value and making sure that every single one of our customers gets the most value for every dollar that they spend. There's an asymmetrical advantage that we have here in the U.S. For example, our existing customers pay less than new customers joining. That's the opposite at our competitors, where all of their existing customers pay more than the customers that are joining.
We feel like we're very well-positioned here in terms of our overall value creation, what we're providing to customers, and how we would be competing in this new world. It's pretty early innings, so we're going to watch closely. We're going to learn. We understand what the risks are. We understand also we're very clear-eyed about what the opportunities are as well. For us, we like these kinds of conversations. We like having more conversations where people can reappraise their existing wireless relationships and conclude that T-Mobile is the very best value and the very best deal that they can get. At T-Mobile, you don't have to make this trade-off. This is a first time recently in the history of our industry where you don't have to make the trade-off between the best value and the very best network.
Having more conversations about that, we see a whole lot more opportunities, but we're also very clear-eyed, and we're going to stay close to this overall environment of what's going to be unfolding here. I'm sure we'll have more to say as the quarters go on.
Okay, next is Carl Murdock-Smith at Citi.
Thanks, Hannes. Quick question. In terms of the collaboration between the U.S. and the European operations, to what extent can you invest together? To the extent that you can not, is it more due to regulatory and legal differences between the U.S. and Europe, or is it more due to the requirement to behave on an arm's length relationship between Deutsche Telekom and TMUS? Thank you.
Let me take a stab at that, and I think maybe Tim can weigh in on that as well from a DT side. Look, there is a lot of things that we need to be working together that does not trip any of the regulatory concerns. For example, T-Mobile U.S. and DT has always had, for instance, a global challenge called T Challenge, where we actually invite some of the best innovators in the world to solve a particular topic. It is like an annual hackathon where we actually reward the companies with cash as well as the ability to roll their products into both T-Mobile U.S. and DT. Recently, at MWC Barcelona last year, we announced the creation of a joint 6G innovation lab, based in Berlin as well in Bellevue, Washington, to actually work on 6G use cases together, like sensing as an example.
Like I said earlier in my presentation, both engineering teams are working closely to influence the 6G standards to drive basically priorities that we believe that is needed for a more customer-centric experience with 6G networks. Those are just some of the examples. There are a lot of things that is happening between both companies that we really value, and it really drives basically the ability for us to use our scale and our global scale and influence to drive a lot of these ecosystem potentials.
Okay. Any more questions in the room at this point? Okay, that is good then. Thank you very much. That was great that you joined us.
Hannes.
Yeah.
Thank you very much. Hannes, are you-
That you joined us.
Are you keeping them there until midnight tonight, or what is going on?
No, we have one more session, and we will soon conclude this, but we are being very thorough. You know this, so therefore, today is another example. You know this.
Thanks, guys.
You bet. Thank you, guys. Appreciate you. Thanks for the time.
Okay. We now move on to the last presentation of today, which is by Christian. Christian is currently our Board member of Finance, but also has been deputizing as Board member for Product and Technology. He is the one who needs to wrap this up. Okay, Christian, go for it.
As we are getting to an end, look, the purpose of today's meeting was to make AI as tangible as possible, to show you what we are really doing, to show you how we think about scaling AI in the enterprise, to get from an MVP point of view to really large-scale implementations. What Tim said is to show you that we have a positive sentiment towards AI, which helps us to experiment with this technology, but also to update our expectation on the benefits of AI. If you take a look at the next chart, you see we were talking about efficiencies, customer experience, and revenue generation today. In many cases, by the way, efficiency and customer experience is not a trade-off.
It is actually adding to another, and the spend pool is large. Over the whole group, we have EUR 45 billion of OpEx, of which 50%, roughly 50%, is personnel cost. In addition to that, we have another EUR 17 billion of CapEx. AI can be attacked to a large degree in almost every category which we are experimenting, and I think that was the purpose of today's meeting. If we are zooming in on the spend pools of DT ex US, you see we are spending roughly EUR 10 billion in the technology space, another roughly EUR 5 billion basically being split between sales and service, but also IT. We wanted to show you today, with our use cases, how we address these spend pools and how we make them way more efficient.
You know, as we said last year, especially when it comes to CapEx, that we reinvest any kind of efficiency which we gain into more rural build-out in the German environment or a full build-out on the MDUs. This is too early. If we are going back to the other categories, which we call G&A, this is actually everything except for technology, IT, and sales and service. It is not a general G&A. It is basically what has been left over. What we are doing here is not that large-scale basis, but highly efficient cases. We use it in financial planning. We have machine learning helping us in this one. I think on procurement, you heard about the efficiency in the fiber build-out, and Alex showed this today in the process. We have a machine learning model which actually connects every construction provider with every geolocation and the demand.
And this model automatically gives you the theoretically most efficient spend, which obviously has to be then approved by the regional technical head. We are using in real estate management, we are using AI in order to steer energy consumption. T-Systems is experimenting with a digital Magenta worker. A digital Magenta worker has an employee number, but it does not require vacation. What they are doing is they are doing monthly account closing. We still have double work, so we have this checking also manually, but we are trying to do, let us say, repetitive work basically done by agents. These are just a subset of numbers. There are so many more to listen, but I think I keep it on this one. Now to the efficiencies. This is what we provided in October 2024, EUR 0.8 billion, of which EUR 700 million is OpEx.
We have updated our model, and we increased it now by EUR 300 million on OpEx, and this is very much driven both in the network space and in the sales and service space. Does that lead to any kind of additional guidance numbers? No, it does not, and there are several reasons for this. First, we are massively investing into AI. Secondly, as we are in the process of scaling, let us be honest, we have not scaled everything, right?
We are in the process, so therefore, I think we are not at the same efficiency level as we are, what I would call on horizontal flight modus. What we have not talked about, for obvious reasons, is there are many projects were trailed, but they still cost a lot of money. Therefore, I think this is why we have not increased the guidance. Can we have the next chart?
Someone has to move the next.
We keep it, and secondly, on the free cash flow chart for the DT ex US business, we are spending the money back into the market, and you know that we have expanded both the financial envelope for CapEx, but also the efficiency to basically attack the alt nets in their territories and to have more build-out, initial build-out in the MDUs. If we are moving forward, can I have the next chart? What is the good news is as we are scaling AI, we expect roughly this EUR 1 billion by the end of 2027, which is roughly EUR 1 billion of OpEx of gross savings. As we are scaling, the efficiency will increase, and therefore, the impact of AI will accelerate over the time. This is why we expect roughly EUR 2.5 billion of gross savings by the end of 2030.
Obviously, this is very dynamic. As we are moving, we are always back-testing our assumptions also in the future. Let us go to the next chart. Can somebody? This is kind of our current understanding, what is kind of the IDC savings we are getting in the technology, in the IT space, in the sales and service space, and in G&A. This will be updated on a regular basis because we do not know what we do not know. Next chart. This is an important one. By the way, all the guys who are spending tokens are almost out of this room here. This is how do you manage token costs, right? First of all, obviously, there must be a model tiering so that we have a clear line of sight what are cheaper models, what are more comprehensive models.
There is model cascading so that you have cheap models doing the work first, and then you have more comprehensive models doing the finalization. Kartik was talking about switchable harness so that we are not vendor-dependent. I think this is something which we are building up. We are using statistical methods that we do not, let us say, exaggerate in reasoning. If it is good enough, we stop. I think these are the things how we basically do it from a design point of view. What is as important is how we operate it.
First of all, all the spend on tokens will be charged back to the segments. They have to feel what it costs, because they are not only generating the gross savings, they also have to bear the cost. Secondly, we can track this on an employee level. We know exactly which employee has used which kind of models.
The second one, what we are doing is, we are doing a pretty flexible vendor management. You heard Tim talking about ChatGPT licenses and Copilot licenses and small models and all these licenses. What we do not do is, we do not give any usage commitments, neither on employees nor on tokens. We have that flexibility that we can switch back and forth. The third one here is important, the tier token budgets. Simple tasks can be done by everyone, and then you get S, M, L packages on tokens, which you can use. Everything has to be tracked by a cost center, a guy. Otherwise, we do not see how we optimize this. The design phase is obviously with the guys who are providing all these models, but the operating phase has to be strictly managed by the cost centers.
Our assumption right now, and take it with a grain of salt, is token costs will not exceed a low double-digit percentage of the gross savings. If it does, then we have to actually calculate, well, this is the best way to basically create efficiencies, because there is also a very classical one, for example, on small processes offshore. You do not have to automate everything. We have to basically compare the efficiency between AI versus classical methods. On revenues, I think we talked about this one here, and this is more tangible. The EUR 800 million, the EUR 200 million to EUR 800 million, which Ferri was talking about, both in the enterprise space, but also in the SME space. What is not included is physical AI. Physical AI, for example, AI Gigafactory, something like this, is not included in this equation.
You see, can we go back, please?
Next time, I need to click. Can we go?
What you see here is what we talked about in the consumer space. It was churn prevention, it was hyper-personalization, it was AI calling assistance. We are a little bit hesitant to call out this number, and James, you called us on this one, because it is hard to retrieve. What is now the AI impact on churn prevention relative to other elements? It will be baked into a total number. The first one, pricing power from better customer experience, I cannot wait to see this. Our industry has not been really great in utilizing pricing power. I hope this is now something which we will utilize. Okay, next one. How do we think about AI-related investments? It is basically threefold.
One is, Ferri was talking about the Tucherpark, the AI Gigafactory. We put this on the balance sheet. This is hundreds of million, but not significant. This one is basically being put on the balance sheet. Then we are considering to pitch for the AI Gigafactory. It hasn't been decided, but then we're talking about beyond 100 MW, and that will probably be done with partners on an off-balance sheet construct, as you know. Then we have a financial arm, which is Deutsche Telekom Capital Partners, who are investing massively with their infra fund into this. You see maincubes with 400 MW or GreenScale were supposed to get to one gigawatt, where we basically just have a financial return. But this is a financial investment, which obviously is also a minority stakeholding, which we have.
Lastly, I think to sum it up, I wanted to be as quick as possible. We increased the efficiencies from 800 to 1.1 without increasing the guidance, but we expect an acceleration towards 2030 to EUR 2.5 billion of the addressable OpEx. That's it. Thank you.
Christian, you stay on stage, and we're getting a few comfy seats for the board members who join us now for the final Q&A. I'm glad we are back down to a half an hour delay. Well done, Christian. Yeah. Good stuff. But numbers speak for themselves, like these kind of numbers in this round. Anyway, I guess, DJ, you join us as well. Very good. I think we need one more chair. No, we don't. No, no. It's all good. One, two, three, four, five, six.
Ferri. Ferri field, yeah. Ferri, come here. The magic seven. There we go.
Magic seven, come on.
Yeah.
Come on here.
Okay. Sorry. All right. Okay, so over to you. I will start with Robert on the left side. Yep. So if we could have the mic over here.
Okay. Robert from Deutsche Bank. Wow, amazing day. Heard all about things I had never imagined I would hear today. You are on the cutting edge. Your targets that you have given, though, is that a case of you just have to implement what you have shown us today? Some of it is already active in some markets, so you just have to expand it across the footprint, or are you have to invent something else or innovate something more to get those targets. Then my second question is, it is maybe one for Tim. Right at the start, you said trust and sovereignty is a very important advantage for you. Do you think you can manage that process within the global telco, or is that an intrinsic conflict? Thanks.
I take the first one? Okay. I would say, I think we have a lot of ideas. We show them today. I think it is about scaling now. I think the whole challenge is how do we go from one country or one POC into real scale? I think we have good learnings from it and I think the money and the efficiency will come from scaling.
I would just complement by saying that we are at the beginning of the journey. I think we are all very encouraged by the early progress that we are making. Every week there is something new. Every week there is a new technology coming out. I personally believe that there is tremendous upside, but also true to our DNA, we prefer to be very careful with what we put in our plans because that is what we understand and know and commit to today. But I can tell you that to the inside, we set the bar and the ambition higher.
Maybe, DJ?
Robert, just to echo what Rodrigo was saying, the EUR 2.5 billion through 2030 is at the upper end of where we think we can land across the four areas. If you run the math, you kind of get to the upper end of the range based on what we know today. In all honesty, if you look at sales and service, network, IT, as well as the other G&A stuff, there is still a lot of wood we need to chop to be able to underpin this with initiatives across the entire portfolio of underlying processes. It might not look aspirational. It is going to be baked into our internal mid-range plans through 2030, and there is still a lot we need to do to get there.
Maybe one more sentence on where we stand. I think after four years now on this journey, it is across the whole company now where we see activities, no question about that one. I would say where we are not really good at is this cross-functional work. Think about, let us say somebody is going as an employee to a hotel and booking a bill. This bill has to be signed off by the boss. The boss then it is going to the compliance department. The compliance department is then giving it to accounting. Accounting gives it then to the HR community to make a reimbursement. We have suddenly 10, 12 different departments involved. In the future, there is an agent in between. The bill is digital, goes through the whole workflow, and you can cut out people across every kind of departments involved.
This workflow design is my next task to go into the big workflows of the organization and to say: How do we reshape the way how we are working in an AI world? This is, I think, something which we have to do. I am very encouraged, by the way, how we work in cross-functionally in Europe. That is easier then to work with the U.S. because of time, distance, and all these kind of things, and we have this classical incumbent structure with mobile and fixed line here and all these things. But what the breakthrough, I think is the UDP. This unified data platform. That is, by the way, why we are not working in stealth modes here. You can even ask, why are we educating the whole market? I can tell you, our competitors are watching us the whole day.
Tomorrow they will say we have that already, everything. I don't believe it because I know it's not true. I can tell you, we have an architecture implemented like UDP where the data is there. Based on this one, we have prompts or agents who are developed locally and used everywhere, and we have these kind of big rocks which we are driving centrally. That is a process, a very disciplined execution, which is taking place in the organization. I think the UDP, the basis that access to data is democratized in this kind of European entities, this is the breakthrough for the leverage of AI in our company.
Very good. Mathieu?
Thanks. Mathieu Robilliard from Barclays. To some extent, you started to address one of the questions I had, which is obviously you're the first one in the telecom industry to do this big AI Investor Day. I'm sure that in the next three to six months, your competitors will do the same. Which is fine, of course, it's great. The point is: How do we differentiate you and where can we think you have a competitive advantage? Is it because you have more scale? You're the largest player, so you can invest more. Is it because, and that's not negative, but you're a legacy complex organization and maybe AI can do more at a company like yours than at a very simple challenger? That's kind of my question.
Go ahead.
Earn your money.
It's okay.
It connects to your former question, how would we differentiate in B2B and in AI and B2B? I think my honest belief is so far we simply didn't ask too many question and we acted. That sounds now stupid and simple and naive, but it's the reality. We would be nowhere if we would have asked for business cases on the Munich factory, for instance. We did it. Did we see the competitors acting like that? No. Now the same goes for this day. Did we see it before? No. You can now ask: Can they follow? Probably they can follow, but we have the advantage of being the first mover. That's my honest belief. We will be also fast on the second step, on the third step, and the fourth step.
To follow up on the question, do I, for instance, believe that we need more innovation to just deliver the numbers? Of course, you need always more innovation. But so far, we also achieved to do everything within the envelope. You have seen the numbers. We didn't differ from the guidance. So we have done all of that. We moved into new terrain, and we didn't harm the guidance. I think this is at the moment where we have a bit of advantage. We have the experience also. Therefore, yes, of course, watch the competition that will never go away, but also be self-confident and fast.
Look, I would say, Mathieu, it's threefold. First, it's the attitude in the organization. We have a positive mindset towards AI, and this is hard to replicate. You cannot buy this from the market. The second one is, you've seen this now across the different presentations. We develop once, but we're rolling out across the whole footprint, so scale actually matters. The third one that's related to Robert's question is, if we address AI in a different modus, that we put the process in the middle and then rearrange the organization around this, that's going to be the big leapfrog. This is what Kartik is always reminding us. Using AI along an existing process is helpful, but redesign the process in a way and then basically put AI at the center, that is transformational. I think this is what we want to address.
I just want to add one sentence for B2C, because I think we have already a few tools where we engage with customers in a very different way than our competitors. If I take, for instance, the way the app or the T-Life, but we have Magenta App, which is quite similar to that. We have in there Magenta Moments, where we have really engagement with customer. The more engagement you have with customer, the more type of different products and offer you can already today give to the customer. In Magenta Moments, it started with deals, but now we have gifting, we will have dining, we will have traveling. So we'll have a lot of opportunities to engage with customer. When you put AI on top of that becomes really something that can scale and become big.
I think we leapfrog there versus a lot of our competitors because of that. Secondly, we use as well our footprints, I think in a very clever way. We do not necessarily have everything in one place. We have different element that we test in different countries, which give us a lot of opportunities to test a lot of concept, see which are the one that scales, and then copy with pride across the whole footprint. I think that's also something quite unique for DT versus competition.
One last comment. I'm spending the last three, four years, 30% of my work time on AI. By the way, understanding how the harnesses are working, understanding how the systems are working, it's controlling all the projects in this organization on the, we call it Leading Digital Tuesdays, where we are following up on that one. On top of that, we have made this big announcement with Jensen Huang on Nvidia. By the way, we had more media coverage with this photo than we had with the iPhone launch. In social media and everywhere. People see that I'm caring about that on a daily basis. People see that we are outside relevant player on this once. On top of that, we are linking AI always to something easy. Maybe not always it is always funny, but at least easiness. Like you saw the video clipping and other things.
I can give you hundreds of examples of what we are doing. This is taking away this kind of burden which people have with AI. They are becoming more Homo ludens, the playful people, and they are trying out. We are trying to enable this playfulness in the organization. This is creating a momentum. I do not have a single situation, even the Works Council, where I was always provoking, saying, "You stop, you make me slow." They are supporting us on all the PSA, the security applications, data security. We always have to consider that in Germany, but it is working. Therefore, I think this is creating momentum. At the end of the day, it is not about the projects. You can find partners' technologies. It is about the mindset and the attitude of people, how they get it done. It starts with the bosses.
I do not know whether my CEO bosses in the telco industries are as nerdy as I am on AI. I do not know, but I do not care. I know that it makes something with me and the others when I am spending so much time on it.
Anyway, we very much look forward to the other Telco AI Investor Days. Greetings to all those who are listening to us. I know a few of them. I think you know. Akhil, please.
Hi. Thanks for the questions. I have got two, please. The first is on scaling AI and just how we should think about it. I guess what I am trying to understand is you have talked a lot about innovation, you have talked a lot about the customer service benefits, the churn benefits, but one of the bits that you have obviously given us numbers to but is harder and maybe not as tangible is the revenue side.
I guess if you look at the telco sector, it has been always one of the challenges of this industry that we have struggled to monetize the opportunities in terms of top line. So I guess I would love to understand how do you think about what it takes for DT to be able to more tangibly scale the revenue opportunities from here, and what are the challenges? Is it because telcos are national, they are not global?
Is it something else? How do you think about that topic, firstly? The second question is a bit linked to that I was quite interested by a common thread amongst some of the slides you had around DTCP, and you are starting to get some benefits from that through some of the investments you have done in DTCP, those assets now being partners and benefiting your AI journey. So I guess I am just trying to understand in a very similar way, what would it take for DT to want to become more aggressive with how you pursue that partnership model to scale it more and help you with your AI journey?
Okay, I start and then you chip in here. By the way, the first thing what I believe is that we have to separate revenues. If we go down the stack in the infrastructure piece, it is very easy. To be honest, if you look to the return on investments, if you look to the leverage capabilities, if you look to the assets, the utilization and the economies of scale around it, we are comparing the same logic, what we are doing for the networks, with the data center logic.
By the way, I have to say, if this is now materializing perspectively, the business of data centers is more economical viable than our fiber investments in Germany. Yeah, so I am not taking any kind of conclusions out of that because we are moving, Rodrigo, don't worry. But it is super interesting to understand the economics of building a data center.
I would love to have take or pay contracts with my fiber customers. The utilization, we don't build in a gigafactory without utilization. We will talk to the large language models or the foundation partners first. We will talk to the government first. We will grow only the data center capacity if we see that there is load coming. I would love to have that for my infrastructure as well. So maybe the terminal value of our fiber network is higher, Rodrigo, so don't worry, yeah? The second topic is if it comes to the upstream, when it comes to the revenue potential, when it comes to additional sales. To be honest, I see all the found encouraging projects, and there is a different way of what Ferri is making with Peter Lorenz in his digital services because we see already the order entries. We see already the money.
Look, the success of T-Systems, which we have seen, and by the way, it is going on in this environment, is driven by AI applications. So it is not something where seeing is believing. You can see it already. On other side, I would say that is my biggest challenge is the qualification of the sales reps and in the SME area. They are used to sell telecommunication services. They are used to sell maybe some adjacency around security and the like, but that they are capable to help customers to integrate their AI applications, and that is why we are making MMS stronger. Remember, we moved MMS into the German organization to have this kind of maybe perspectively FDEs, so forward deployment experts who can help customers to develop this solution, but we have to qualify the things to materialize these revenues.
I can maybe just add to what Tim was saying. I think at some point in time, it becomes incredibly hard to start disaggregating, for example, churn benefits that you can attribute to AI versus other stuff that you do, improving the product, investing in the network, et cetera. Given the scale of the efficiency opportunity, and we have for the first time given you a sense of the disaggregation of our cost base across network IT, et cetera, I think it would be negligent on our part if we weren't addressing that first and foremost of all. Not to say that we need to stay sharp on the other opportunities.
On your second question on DTCP, and I will start and please do chime in, everyone else on stage. First of all, there are three ways we engage with DTCP. On the first side, we are passive investors in their funds. Yeah, so maincubes, GreenScale. We have a commitment to the so-called Digital Infrastructure Vehicle in DTCP, and we learn from that process, yeah? Then we use the DTCP team in two very distinct fashions. Firstly, to assess investment opportunities in growth equity and early-stage venture, which we do on our own books.
This isn't an off-balance-sheet investment in Nscale or n8n or Black Forest Labs, et cetera. That's very much capital from the balance sheet. It's small money. But we do that, and we tie that to rights that we get from these companies in terms of observing what they do, their roadmaps. In some cases, we have board observer seats. And it's-
It's an X fund, right?
That's the ventures piece. The third piece is they also support us as a de facto investment bank, for the lack of a better word, to make investments on the security and B2B side that are synergistic to our core business. Could we do more? Potentially. But just to give you a sense of what we do across these three angles, we have a very tight collaboration. I'm on the investment committee for both the tech fund as well as the ventures IC, and we convene on a monthly basis to make these decisions.
But to make that concrete, I do think never before the collaboration between the investment arm of DTCP and the strategic arm worked that close together. To give you some examples, HR robot, Quantum Systems, Greenfield, maincubes, those companies we always take in consideration. So we have them first on our AI factory. They take load. We invest in them. We do drone business with them. We do robot business with them. So really that's in the meantime a complete different collaboration. If I would not have the chance through Vincente and the team to invest early in things like drones, I could not go up until the stack and vice versa. So we can combine them into our factory. We can then data center capacity, we can have people. So I think this machine is just about to start rolling.
Just one addition. When we talk about revenues, we focus a lot in the industry in general on what is the next, the new revenue monetization opportunity, and that is good. We should also think about the EUR 40 billion of revenues that we already do in our core business. I think that what we presented to you today will help us with a further differentiation. I think it increases the differentiation, and with that differentiation comes pricing power. I think I am already excited by what I am seeing on the fiber side, for example, in Germany, how everything we showed today gives us more speed and gives us more efficiencies and give us more scale and how this virtuous cycle works. I hope that as an industry out of this, for sure we will do it, but we regain some of that pricing power that we lost.
I definitely think this technology plays in our favor.
In general, we are open for partnership models. I think our industry only works in partnership models. By the way, I was open even to apply for the Gigafactory with Schwarz Group. I am open to partnership models because I think the cake is big enough, and it makes it more efficient. We are open for large language models as well because they are asking us for distribution. I can tell you they have great models, but their biggest problem is how can they monetize it? Ferri said it very well. It is like the profit is up there, so they have to go up the stack. The large language model alone will not make the cake. They have to go up there, and then they need the access to the customers, then they need the trust, then they need these kind of things.
Therefore, we are very intensively discussing with some of the foundation models about a deeper partnership going forward. I think this ecosystem is respecting us. We are anyhow in their footprint, number one, which is the Western world. This is where we can offer something, and we are the two biggest economies of the Western world, which is U.S. and especially Germany is very interesting for them. More to come. We are working on this one, but this is definitely an opportunity.
Okay, we can take about one more question at the current pace of our. No, it was a joke, Tim. I once said at the current pace of our answering.
That's true.
We can take about one more question. Okay, we continue with James.
Thank you. Yes, hopefully, this will be a quick question. Well, firstly, I want to thank you so much indeed for everything you've put on today. It's been super interesting, and the vision you've set out. And within that, you have suggested that your gross indirect cost savings will actually, the rate at which you generate them should accelerate from 2027 to 2030 versus the current pace. From that, can we infer that you expect the rate at which your margins will grow overall will also accelerate from here? If you're not willing to commit to that, because I can see Dan and Jay smiling already, what's holding you back from not being able to give that commitment, given the kind of confidence you see in AI today?
There are two reasons, James, why I would be hesitant at this point in time to suggest that there is an immediate tie-in to margin expansion. We would most certainly strive for that, but there are two imponderables. Firstly, inflationary pressures, so wage inflationary pressures. We make some assumptions around IMF forecasts going forward. You tell me whether they are good, bad, or ugly. That is number one. That is an offsetting effect. Then there is a second aspect which Christian alluded to, which is the cost of running these AI systems. I think we are starting to wrap our minds around how we can manage/mitigate some of those costs, but that is an imponderable.
I think to be quite frank, it is premature at this point in time, given where we are on this journey, to say that we would think that this would result in 100 or 200 basis points in margin expansion that we can solely attribute to those initiatives. But rest assured, that has to be the ambition. Otherwise, we are effectively spinning wheels or running hard to stand still. Call it what you wish.
Yeah, I think we can probably safely say that IDC ratio is expected to improve because one other, we are talking today about our indirect costs. We did not talk about our direct costs, and there are components of the direct costs that are very unpredictable, which is why as a target KPI, we have focused on IDC AL and I think on that ratio. On that ratio, I think we are quite confident that we will see this acceleration.
By the way, I am anyhow believing that what we are doing here and the opportunities, if you do just your human intelligence and taking all the measures together, seeing the opportunities across the whole value chain, seeing the transformation going forward. The question is how fast and how consequent we are able to execute. By the way, even transforming our workforce in Germany and other places, that this AI has a super high upside on the world in which we are operating. But now looking to the current situation where we sit with our conviction in the market with regard to our numbers for the multiples and like, it would be crazy to overcommit now and to make additional commitments on this one. We will deliver on the Capital Markets Day. We will, let us say, do everything which we have laid out.
You have confidence that we have a lot of, let's say, instruments how to realize that. I am much more bullish on the future on this company going forward, knowing that there are a lot of, let's say, dependencies which might hit us. But if you put the math together on these measures, there is definitely a lot of opportunities for telcos with their mass-market services, to take benefit out of AI.
Okay. Next we have Ulrich.
Yeah, thanks. I was at Mobile World Congress, I think it is called. Nokia and Ericsson were essentially talking about this traffic change, traffic pattern change of more uplink. They are talking about that as their next revenue cycle for obvious reasons, to their investors. I am just wondering, you already sort of evaded the question of the AI data center capital allocation a little bit. Is there a world out there, obviously beyond the current forecast kind of horizon, which you have nailed down, and then there is no debate about it, but beyond 2027 where, because of AI and the demands it puts on a network, the capital intensity of the business structurally rises? What is capital intensity?
I can answer it.
Yeah. I correct you.
No. Generally, the answer is no, but maybe we want to.
No, but to be honest, the question is how we are managing it. If we do it in an off-balance solution, the answer is no. Because, then we take the benefit from the revenues and from the upside on our equation. But if you include the off-balance capacities into our balance sheet, then it's following the same logic, almost the same logic. By the way, the capital turnover of data center is higher than our fiber. Yeah, so it's slightly better than our fiber investment, which is very costly, as you all know. But I think the question is heading to our hypothesis. Look, we want to grow our return on capital employed into double digit. That is our ambition, which we have formulated very clearly. We want to utilize the infrastructure in the most efficient way going forward.
We have a CapEx envelope. We have a leverage ratio where we have said 2.5 is, let's say, the God-given threshold which we follow. In this envelope, our industry and our company will evolve. I am not now speculating that I am going now in a 200 billion investment cycle like the Americans are doing. This is impossible for a telco. It is impossible for my business model. Any other question, how all the money is going into this industry at that point in time, and how much of this will be diluted later on, I do not believe that all of this money makes productive money. But it is not our strategy.
If you followed what Ferri have said is, we are very much following a strategy of sovereignty in our data center logic. We are very much following a story of, let's see how we get utilization first, take or pay. We cross the rivers by feeling the stones. That is our story today. We are very close to governmental and security services and the classifieds here and developing that with them. With the industry we are merging with RAN.
We are not going into a kind of trip here by saying now all in, and then we will see what is happening. We have to do it this way because we have no large language model in the back, which is automatically feeding this beast. That is the way how we are doing it. We are not competing with hyperscalers. That is not our ambition, but we see a big opportunity for a market in which Deutsche Telekom sits at that point in time with geopolitical tensions, and which we want to try to monetize.
Why not?
Very briefly, just alluding to, let's call it the core business, the core telco business, because I think your question was more around CapEx risks associated with the core telco business in the medium term. Maybe a couple of things. First of all, we had guided to roughly a 21% CapEx intensity ratio at last capital markets. We are trading slightly above that, but that's well telegraphed because that's an investment in fiber, specifically in fiber Germany. The second point is, we are in the process currently of upgrading our mid-range plans. I cannot recall a single conversation in which Alex Jenbar or Arash or somebody else approached us and said, in addition to whatever we currently have on our roadmaps for network modernization, and I mean RAN modernization, that there is an incremental requirement for an investment to be able to prepare ourselves for additional uplink/AI-related traffic.
Just to give you a sense of the degree of confidence we have in terms of our own CapEx planning, specifically related to mobile network capacity. I think those are the two things that I just want to add to the mix, and all of the other debates on, I think Tim's already mentioned that, so I would be wasting people's time.
That doesn't refrain us, I must say. When I am sitting here, I like that, and then Tim did start the race on AI, and he allowed us really to think wild, and it was good to stay within the envelope, as it really keeps us with the eyes on the street, and we really went for utilization, and that meant we went and stayed with the customer. So we didn't just fantasize around, and I liked it.
Now let's imagine for the next steps, the next stones in the river that we can cross. Trust to us also means in terms of industry and customer, public defense health. If you now combine, for instance, what's going to happen together with the Ministry of Digitalization. And in that regard, w e came through cloud into the play to help Karsten Wildberger on that. If you look on the defense side, we are a strong supplier for the Bundeswehr and others. We are now together with DTCP in the drone play, in the robot play. We can go here in big steps, which still are stones in the river, which still keep us within the envelope, but these are huge steps.
Okay. David?
Yeah. Echoing everyone's comment, thank you very much for your presentations today. Very helpful. I am listening to this and I am understanding the complexity of the investment in networks, how deep the intelligence is moving into the networks. I am just thinking ahead, and I would appreciate your views on this. Does the shape of the industry change with AI? What I am talking about is should we have four mobile networks in the market? Will we have four mobile networks? Will we have a big fiber operator, a second fiber operator, smaller fiber operators as we go through fiber investment, 5G investment, 6G investment? It feels like things need to change a little. Is that the right interpretation?
It is a complex question. I think you allude to the fact that probably fixed operators have some advantage in providing uplink and low latency in mobile as well, because we sometimes read a lot that we use our fixed footprint as a kind of edge place. I think that is true. I think there is an advantage that we have in Europe doing that. I think it is very different in the U.S. because they have way more spectrum and things like that. I think in Europe, we have an advantage of having both because you can download, you can use some of your fixed infrastructures to provide more better latency and edge computing.
But look, your question was more about market structure and CapEx intensity in the markets. I think it is market by market different. I think there is no rule of thumb. But what I see definitely is first, and we see that already today, there are huge productivity gains if you deploy AI into the way how you are rolling out infrastructure going forward. Yes, there are some productivity gains. The bigger you are, the higher they are. That is an advantage. Second, I anyhow believe that this industry is under super pressure. And by the way, I feel under pressure. The whole industry feels under pressure. And we discussed it endlessly. There is that 60% of the industry is not earning their capital cost and all these kind of things. I anyhow believe we are already in the middle of the market consolidation with regard to infrastructure.
And if you ask me, do we need four infrastructures? No, we do not. We do not need four mobile networks in Germany. And by the way, they will never take place. I make with you a bet here, and I did it already. Because from a capital efficiency, it does not make sense. Therefore, I think this consolidation of infrastructure is happening, has happened in Spain, has happened in France, it has happened in U.K., it is happening in Poland, it is happening in Hungary. I can go on and on. We are in the middle of it because the capital efficiency is something which is that we cannot increase the prices endlessly to just get a monetization. We cannot expand customers because they have already high saturation of certain services. Therefore, the only way of doing it is by consolidation and by bringing less to it.
That is, by the way, why we even see some MVNOs rising and trying to help these companies to get contribution to earnings. Therefore, I think your question is simple. Yes, in five, 10 years from now, there is a market logic that this is taking place. And on top of that, I can tell you one thing. I was with Commissioner Ribera, and I said, "Oh, it is great that you invite these big telcos here." And she looked at me. "Why you are so important?" "No," I said, "In 10 years, nobody will invite us anymore." Because this group of dwarfs, which are sitting here in this world of the AI ecosystem, are not relevant for Europe anymore in 10 years from now if we are not doing something.
And therefore, I think political leaders understand more and more that this fragmentation of the telecom market, this kind of atomization, these real estate companies which we sometimes see, this is not a surviving business model going forward. Therefore, you know my hypothesis. Consolidation is taking place in the infrastructure to gain the efficiency on that one. And to be honest, I do not care about these players. I care only about Deutsche Telekom, and we have the scale in our markets to play out that card and to be the survivor in this environment here in our markets. That is at least my assessment of how Europe's markets is evolving.
Well, good. Emmett, I like it.
Yeah. Nice one. Cheers. Thank you. It's Emmett Kelly at Morgan Stanley. Tim, Ferri, I think the answers you guys gave on building data centers and data center CapEx, I think it's very clear. If I kind of put you into a global context and compare you with some of the other big telcos I see around the world, they're looking at things a little bit differently. I know SK Telecom is a company, Tim, that you know and you admire them a lot. You've signed partnerships with Singtel. SK Telecom, I think, is rolling out over 10 gigawatts of data center capacity. Singtel is really going for it. I think about Xavier Niel in France. He's going to roll out over 1 gigawatt of data center capacity in partnership with InfraVia, and there's other telcos doing something similar.
Why is their strategy, do you think, so different to yours in terms of the data center spend? Do you think they're more risk on? Some of these telcos are slightly redefining themselves as tech companies. Is it a business model question? Why do you think the big difference? Thank you.
The answer is Christian. No, I'm kidding. Look, by the way, I admire very much SK and even Singtel, we are very good in intense contact. I understand the way they're coming. There is, I think, one thing, they were earlier in a saturation phase of their telco markets, so they were earlier thinking about how they can generate new kinds of revenue streams across their businesses. Second, their software and tech capabilities are higher than ours from the legacy which they had, so they had a big basis to that one. On top of that, I think, their CapEx needs. They could feed that. We have seen the same thing in China.
I am always going to the regulator say, "Look at China." When I started in this telco industry, becoming a board member, that is now 20 years ago, I can tell you, China Telecom, the Chinese operator, were nothing. Nothing. We were even looking down to them. Today, we are looking up to them. They are the first ones. They have built huge data center capacities. By the way, a relevant market play in this one. They have tokenized their infrastructure, so they are offering already AI in the networks, as we have discussed it this morning. They are trailblazers in telco industries going forward. Therefore, look, we are in an investment hype today. We cannot afford now to go big in data centers. We would kill our investment hypothesis, which you have in us.
Therefore, we will prove you that what we are doing step by step is a nice business to build. But if I would now have a carte blanche and unlimited resources, we would seriously look deeper into the question about building a kind of second pillar around this infrastructure business, which we very well understand. It requires a little bit more than just GPU as a service. It requires a cloud, it requires security requirements, and all these kinds of things, so you have to build the competencies around it. But I can tell you, it is a very attractive market to invest into. Therefore, we are in our framework. I think it will help us going forward. It builds a lot of credibility for our B2B customers that we are in the data center environment. I think Ferri made that clear.
Now we see how we are developing this over time.
DTCP, by the way, the DIV II fund it is building about a gigawatt of infrastructure.
Yes.
And we have a 36% equity stake in it, so it is-
Yeah.
Not nothing. It is not nothing that maincubes had a capital raise of EUR 2.5 billion earlier this year, so that is a proper size. So we have exposure. I think we are at the end, unless there are any more questions? No. Tim, do you want any
Yeah, let me give you a 15-pages presentation to wake you up again, friends. So look, very quickly, by the way. First, I would like to thank my team, my board fellows here, colleagues, but as well, the whole team who have supported that work. It was a tremendous work of what they had done, and I think it was worth doing a job to just bring it up once into a good piece of paper. Thank you, everybody, for doing this good job. And please, messages even to the colleagues who are not here anymore.
Colleagues.
The second one is, I asked you at the beginning three questions. The first one, can DT execute AI at scale? What would you say? Yes? No? Think about it. Can we capture real value from it? Second question. And the last one, how will we measure progress? To be honest, I believe this one is the most challenging one, Dhananjay, so big job for you to really, let us say, make that transparent to our investors that we are participating on this iGrowth, and we can see that. I think we have audacious targets. It is not that we are sitting here and just hiding. We have clear targets. And by the way, guys, as we did it from the first capital markets days of this company, this is more about insight than talking to you guys.
Because I can tell you, when we state ourselves personally to the investor world, to the shareholder world, this is a commitment, and this commitment is seen from all the people in this organization. We steer the company from the outside into the inside, as we always did it with the capital markets day targets, and as we do it now again with the AI. This is much more than just doing a nice investor day with you. This is a big commitment which the organization, and I know that from 20 years in being in this leadership team here, that I know that the people are taking our statements serious. That is why it takes so long to find them, but the moment we once agree to this one, this is a plan. I think we have a how. For me, these postures are very important.
Data, we have the UDP. I think this is the underlying basis for everything that we are doing in. If you don't have the data layer, forget about everything about AI. We have it, we develop it, and we have democratized our data. Second, we have deep partnerships. We are not going with one into the bed for life, like I do that with my wife. This is here, we are more open-minded in this regard here. But we will collaborate with the best. I am not worried about it. We have the size. We have the respect, we have the competence that we can partner with them.
I can tell you one thing, I can guarantee one thing, nobody in the Valley looks at us and says, "These idiots from Deutsche Telekom." Maybe, "These guys taking us serious," and they see an opportunity to do something with us in the European hemisphere. The third one is we have an operating model, which we are. I said it, we are at the beginning. We have to go into the workflows, into the processes of this organization. But we have defined the big bets, and we keep you posted on these big bets coming forward. I can tell you, maybe you have seen that among all these cases today, but on sales and service, we are super advanced.
If you don't believe it, do me a favor, spend two hours and go into the service center and just sit aside to an agent and watch what he is doing and how he is working. It is a total different way of how we worked a few years ago. A flexible AI stack, this is very important. By the way, as we said, we have switchable harnesses across the organization. For instance, we use different models behind that, and we are always switching that. We had Glean, for instance, which we used originally on our AskT. It was a little bit too expensive for us, so then we switched it to an open source model. This is the way how we orchestrating the cost of our tokens in the organization going forward.
But even more important are the evals and the sandboxes and the way how we are trusting and testing these functionalities. I found this very impressive, how we see that in the network this morning, how we control MINDR. In the way how we are steering that not suddenly the agents are taking over and switching off our network, which would be the super catastrophe. Therefore, I think these evals and models are really important, but make sure we know how important trust is. The T stands for trust. The last one is human in the lead. We want to make AI not just a technology, not just a model which is a technology which we use. We do not want to substitute the character and the brand value. We are staying a human-centric company.
By the way, I do not believe that all humans are get substituted by AI in the future. We are just sitting there and watching TV or whatever. There is a lot of work which we can do and where we can use and reskill our people in the way where we are. Having 30% digital expert in this company, why not 40, and making a business out of them by saying, "Go to the customers and help the customers to make businesses." This is, I think, the way going forward. Please, I hope that you got a feeling how serious we are. Please talk to us and challenge us on the progress we are making. I hope that you got a feeling that this is a new business model which we are building and not just a tool. This is in our strategy, the accelerator.
Global scale is the one thing. Maybe we should do a capital markets day on global scale. All we do and we do it on AI, but AI is the one which we showed us today. My recommendation going forward is here you see the new board of 2030. Yeah? With all these nice chips. My idea is to develop an AI avatar that I do not have to be physically anymore here, but that my colleagues are sitting there and consulting me while I'm being an AI robot. Was that a good idea, guys? You can switch me off whenever you like. With this, I like to switch off our session today. Thank you for your time, and now it's time for a drink.
Yeah. Thank you, Tim, and thank you everyone who has participated. Thank you for your attendance and good questions. I would like also to say goodbye to those who have followed us in the stream. For those who are here, as Tim has said, there is food and wine, and other drinks around the corner.