Great. Thanks everyone for joining us for the next session. We're delighted to have SAP back at the conference, and particularly Walter Sun, who's the Chief AI Strategy Officer. Walter, I think this is the probably third year in a row you're coming back, but we live in interesting times.
On the kind of AI front, there seems to be almost an exponential curve in terms of the development. I wanted to sort of kick off and get your perspective on, in the past 12 months, what have been the really big step changes that you're seeing? Obviously, the discussion around the application layer has been obviously the most significant. First of all, just set the stage of what is kind of exciting. What have been the biggest advances and how we see that sort of evolving moving forward?
Yep, great. Great to be back, thank you for having me here. Also, given the volume of requests we have coming in from the investor community right now towards me around AI and our team, hopefully, we can address a lot of questions here en masse. That's a great opportunity. Yeah, it's also a good opportunity to reflect. I think when I was here in 2024, first time, you know, generative AI was still kind of new, right? I think we had released around 10 generative AI use cases in production. We had announced Joule, we had announced SAP AI Core, Generative AI hub for building extensions. Fast-forward to last year, we had around 130 generative AI features that we had released.
Joule had around 1,400 skills in it. A skill is basically mapping a user intent to something that Joule can do, right? Like, show me open purchase orders, these kind of things, right? We actually have to map that in the system to the APIs and all the data tables and things like this. And I had spoken about some of the innovations that we were working on then, right? That were exciting me, right? We had three buckets. One was agents. We're just starting to talk about agents. This year we've now released agents, right? We have 30 agents released. We have AI Agent Hub that we've released as well for governing agents, right? We have Joule Agent Builder to extend agents.
Within a span of a year, I mean, it kind of went from where the whole community was kind of defined what an agent even is, and is this feasible to actually have in, like, productive agents? Which moved very quickly. The other area that I talked about was knowledge graph. This is this Neuro-symbolic AI, right? Where you codify and you represent knowledge, and then you provide that to the AI so that it can do its job, basically. We had built a knowledge graph internally at SAP, which is massive.
Just in the ERP system, just so you have a feeling for how complex these things are, it's 452,000 tables and 7.3 million fields in the table, and 80,000 analytics views, and the knowledge graph basically links all of these things together, right? If I'm talking about a sales order, I know which field and table to look in, and which that's linked to which business process and all this kind of stuff. That is now also being used and leveraged by Joule, so that is now productive. You can extend it on the Business Data Cloud, so that's live. The other innovation that I had spoken about back then, were, let's say, alternative foundation models to large language models.
The one that we were building on, and we were building back then, and it was still somewhat experimental. Back then, we called it the SAP Foundation model. And basically, what it was doing is a foundation model built on tabular data to do predictions. So regressions, classifications, like numerical calculations, which large language models cannot do. We didn't really know if it worked back then. Now it's productive. We've released that. We called it RPT-1, so Relational Pre-trained Transformer, RPT-1. That's productive. We have one big chemicals company who is. Christian disclosed this in the earnings call. They have around 180 narrow machine learning models that they have in production.
Things like auto-filling a sales order, predicting delivery dates, these kind of things. They're gonna replace all of those with this one model because you don't need to create a data pipeline and feature engineering and train all these individual machine learning models anymore, and they get higher accuracy. I'd also, and I'll just one last plug: it won, or it was honored with a Spotlight Award at the NeurIPS conference last fall, actually. It's kind of like the top A-star data science conference. Showing we can play up there where we need to with the big AI acquirers. That was the stuff, certainly, I think, over the last year, right? Looking ahead, yeah, I mean, look, it accelerates.
I mean, there's so much stuff going on. I'll, I'll plug the article that we published a couple of weeks ago. We always like make, like predictions, like top five themes. You can look that up, so it's AI in 2026, top five themes. But yes, again, certainly these foundation models that are not large language models, so things like world models and robotics. We have some partnerships that we've announced there. We work with NVIDIA on these kind of things, the tabular foundation model. As well as agentic governance is going to become key. You're releasing all of these agents into the organization now. They need, you know, to borrow from HR, they need a hire-to-retire life cycle.
You need to discover them, onboard them, monitor them, govern them, offboard them, right? Have observability, give them access rights, all this kind of stuff. We've made some announcements around that as well. Sovereign AI, I think is an increasingly important topic, we could see the discussion with Anthropic and the Department of War, I guess they're called now, right? These kind of things. Yeah, I mean, it's, there's just a lot going on. It never slows down. I kind of hope it slows down a little bit sometimes 'cause I'm just exhausted sometimes trying to keep up with it all, as you are as well. It's also really, really exciting and full of opportunities.
Jumping on that point in terms of the pace of it, there has been an accelerated debate around, like, how agentic AI might disrupt application software.
Yeah.
What are your thoughts on the current state of play? Like, who, in your view, would be, like, emerging as the relative winners or losers?
Yeah. The elephant in the room, right? Is SaaS dead? That's more or less the question, right? Big debate right now. No, I mean, look, I think it's two things, like, and, from SAP's perspective, I mean, one is we acknowledge that this is a disruptive technology, and we're leaning into that, and we're going on the offense in certain areas, right? AI will certainly disrupt the user interface, right? How you interact with computers. You know, you can have intent-driven ERP systems where you ask the system to do something, right? You're giving the intent, the prompt in, and then it's going to go and execute and do something, right? You may not need to click in the UI anymore, right?
Also, you know, user interface will increasingly contain elements that are generated. We announced this. We call this Gen UI, Generative UI, and we already have elements of that in Joule, actually, right? If you're pulling up a chart or something like that, it's actually, like, generated on the fly by the application. In the future, we also may have, you know, instead of humans talking to SaaS, you may have, you know, agents talking to SaaS as well, right? Where the human is merely notified. You know, the UI will be disrupted to some extent. You know, I do think UIs are still somewhat sticky, though.
My wife's friend works in a trading company in Zurich, and she was complaining to me 'cause, you know, they've moved off their ECC system that she'd been working in 10 years, and she just knew where to click and just loved it, right? I mean, these things will always coexist, right? You always have a UI, again, the logic of that can still be accessed by agents and Generative UI and all this kind of things. We're leaning into that. We have Joule already rolled out, right? We have a solid base for that. Secondly, you know, this AI will disrupt how software is built and maintained.
Mm.
Right? Certainly. We see how quickly low-code and AI pair programming is progressing, right? That opens up a few opportunities. One is, you know, I mean, I think that's driving some of the fears of disruption as well, right? Now, for us, we also use low-code, right? We've rolled that out to all the developers in the company, and we've seen magnificent improvements in productivity to using that. Now, for us to live code something, you know, it goes through our 388 product standards and testing before it goes into production, right? Okay, it's like enterprise-grade software, but we can live code, and we can also offer that to our customers to quickly build extensions, right? We have SAP Build, for example.
You know, we've released Joule Agent Builder. You know, as a user, you can go in and actually vibe code build extensions to SAP systems, which in our point of view, just increases the value of our systems. The third area, I think, is that, you know, AI can also disrupt the commercial model somewhat. Okay? Right. There's this big debate like, you know, around what's gonna happen with seat-based subscriptions in the future and licensing. Couple points there for, so from SAP's perspective, One is, the current commercial model that we have for artificial intelligence is actually based on consumption.
Mm.
We talked about that a couple years ago. It was a lot of heavy lifting internally. We built out a capability to actually meter every single one of those agents or use cases that is being run in the system. If someone goes in and processes an invoice using AI or a delivery note in the shipping yard or something like this, that's metered.
Mm.
Charged against these AI Units, right? We basically have a commit to consume type model for that. You subscribe to a certain number of AI Units, and then that's pooled, and then you consume that, right? Every AI Unit is tied to a business outcome. Somewhere around, like, messages and things like this, but, like, where we can do it, we try and tie that to, like, a business outcome. Okay? We actually have that already there, right? That's actually already in place for us. The other thing too is that I think it's less than 50%, I'm not sure exactly what we disclosed, 50% or 40% of our it was seat based.
The rest of that is actually based on, different business metrics, like spend under control, for example, business documents in the system, these kinds of things. There, we're going on offense, and then I also think there's defensive moats that we have, right? One is certainly the data that we have, and that's a bit of a truism, right? Like, if you own the data, but it's not just the data, it's owning the data models itself and the contextual meaning around all that data, right? Again, we're exploiting that. We've built the knowledge graph. We have Business Data Cloud now, right? Where we can expose this semantically rich data in SAP and non-SAP systems to be consumed by agents, for example, right?
We have that data, customers trust us with it, and that's the other defensive mode that we have is the customer trust, right? That I mean, you know, probably the biggest argument I've heard against low-code is, you know, it's Friday night at 8:00 P.M., and you have to close the books, and your general ledger doesn't add up. You know, who do you call, right? Is it your SaaS provider to fix that? Or is it the guy who live coded something, right, and tries to figure out what's happened there? There's a lot of customer trust that they have, obviously, with our systems.
I mean, I'm biased, but obviously, I think, you know, we're in a good position to emerge as a winner out of this whole thing.
There's this sort of view that horizontal software is particularly more exposed, you know, to some of this risk than vertical. Even though you could argue that instances of vertical software use cases also getting challenged. You guys are interesting because while you sell horizontal software, you know, you also build for specific industries.
Mm-hmm.
Obviously, manufacturing is a key part of your end market. I mean, when you look at sort of, say, analogies of what happened at the time of the cloud and what's happening with AI, how should we think of kinda your evolution or your ability to kinda navigate this cycle? You know, you talked about the data mode that you have, this process knowledge. When we think about the disruption of this application layer that's happening, you know, the modules as we know them, the line of business applications are clearly changing. So ultimately, from your standpoint, you know, what are the key things we need to look for SAP to kinda navigate? Is Joule ultimately gonna be all end all?
Ultimately, for a lot of customers, they may not want to use Joule. They'll say, "Well, I'll use another agent," right? Help us kinda navigate that kind of blurring lines, right, between that sort of application layer in an AI world.
Yep. Yeah, I'm hesitant to put us in the box of horizontal player, totally.
Right.
'Cause, if you look at what we do in finance, for example, it's quite deep, right? If you wanna know how to process an invoice in Brazil and be compliant with Nota Fiscal and things like that, I mean, we're there. So we are quite deep, and of course, yeah, we have a lot of industry solutions as well, like a customer activity repository in retail or, these kind of solutions. And it's interesting, too, I mean, there's some apps you would think might be disrupted, but they have, like messaging apps and stuff. I have a policy of not talking about competitors and stuff, but where they have, like, network effects, and they seem to be pretty resilient to disruption.
I think if I look back at SAP's history, right, you know, was it 2016? I think we had Clayton Christensen on stage at our big SAP Sapphire event, right? The Innovator's Dilemma, before he passed away. You know, we have a strong history of actually, like, disrupting our own business model, right? Going back to like, you know, moving from client-server architecture, you know, in-memory database, right? Moving to cloud and SaaS, like, that was basically disrupting a nice, profitable business model where we sold licenses and then collected, you know, 20% whatever maintenance on it, and the customer had to do all the installation stuff. And I see that same spirit here, to be honest, right?
We're leaning into the opportunity that generative AI brings for us.
Just wanted to see, like from your lens, how do you see the landscape evolving overall in this agentic AI world? Like, in between the incumbents, the new age players, which would be AI native players, as well as these LLM providers, and also, like, even companies in sourcing to an extent.
I'm thinking of, like, Porter's Five Forces model right here, right? sort of substitutes and rivalry and whatnot. No, look, I mean, we see the models commoditizing, right? That's been a trend for a while now, where the price per token is just falling, and they're pretty much collapsing, right? Converging in terms of capabilities, it's no surprise then that they, at least the model providers, try to move up the stack a little bit. We see them making their first, like, PaaS offerings, right? The OpenAI Frontier, for example, which Gartner, by the way, issued a note two weeks ago, just telling their customers to be careful of it because they probably can't scale it, just repeating what Gartner said.
Cloud CoWork, these kinds of things. Of course, they start to push up the stack a little bit. From our perspective, you know, the hyperscalers and the PaaS vendors have tried for years to creep up the application stack, and I mean, pretty unsuccessfully, to be honest. I think there's a lot of, like, I wouldn't underestimate the knowledge that's needed to build those applications. From our perspective, we treat the SaaS, sorry, the model providers, right, and hyperscalers as we always have, right. We treat them as partners. We actively participate in the A2A protocol, for example, and MCP. We're, like, founding members of MCP protocol. We collaborate with them.
Of course, they're gonna try and creep up the stack a little bit. You know, we do see, obviously, you know, there's around 600 startups I think that we look at right now, in the AI native space, springing up. You know, I think they have probably the challenges that a lot of startups have, right? Is like, just having this, like, enterprise readiness, to be able to compete with us.
Just a follow-up on that, like, how do you think, like, incumbents will look like in terms of strategy? Will they look to?
... these AI native players, or will they try to build all these AI capabilities in-house? Where will the balance be more-
Yeah.
like, heavy?
I think Salesforce and ServiceNow went on a bit of an acquisition spree the past 2 years from AI companies. I don't think the market necessarily rewarded them for that, to be honest. You know, we made an acquisition of WalkMe last year, right? In the HR space. That was just to boost our recruitment module and SuccessFactors. I mean, for us, it's tricky right now because this stuff is pretty new to everybody, so we haven't seen a startup where we say, like, they have some amazing capabilities that aren't available somewhere else, right? We're all kind of cooking with water, and the valuations are insane at the moment on these startups.
I mean, from our perspective, we monitor them, right? We work closely with them. We still partner very broadly with a lot of startups, but we're just kind of like watching the market at the moment.
Got it. I wanted to come back on this sort of comment you made around the kind of data mode. I mean, there's a system of record, right, which is sort of the single source of truth. Pretty kinda hard to displace, but increasingly what we've seen from the likes of, you know, Cresta and, you know, OpenAI's Frontier, that they wanna kinda sit as this sort of abstraction layer on top of that system of record. You know, in the future, is that a kind of a viable option of a direction that we will go into, that they just kinda sit on top, but eventually... You know, 'cause data access is also another key, kind of, important characteristic here, and license agreements, et cetera, play a key role in this.
Is there a scenario where the system of record becomes a kind of dumb pipe? Sorry to use such a strong word, but, or will, 'cause you've got the business logic, the metadata, all that as well. I'm just curious, kind of, is this the next big battleground?
Yeah. Everybody wants to be that abstraction layer, right, and orchestrator sitting on top of other applications, right? I think in reality, the future is going to be, the big players are gonna have their own agent, their own, their own copilot, right? That's doing the orchestration in their systems, and that's gonna have to collaborate with other orchestrators and other agents, right? Joule, for example, again, can access data and context, and authorizations, and metering, and logging, and all this kind of stuff in the SAP systems that a third-party agent simply cannot access, right? To be blunt, and I'm not sure I would want, I'm not sure our customers want us to allow just unfettered access to the SAP systems from some third-party agents because, I mean, that could be an...
This is gonna be like a security and an auditing nightmare, right? Again, I see a future where you'll have these assistants, copilots, right, that are good in their domains, and through protocols like A2A, they will then collaborate together, and I don't see anyone really being able to take over, like, as the Uber orchestrator. I will say as well, we're quite open, we were the first company. We have a bidirectional integration with Microsoft Copilot, that is now GA and Joule, right? I can be in Copilot and say: "Help me book my trip," and that will go over to Joule and Concur and then make a entry in my Outlook calendar and vice versa, right?
We are open in that sense. I don't see anyone realistically taking over. You know, it's it sounds funny because, you, I mean, like Anthropic and OpenAI, they almost have, like, dumb intelligence, right? You know, you can hire your smartest friend to go work in an SAP system, and they will have no idea what to do, right? All that, like, knowledge of how to run a process is, like, codified in our systems. That's what makes it valuable.
Right. Just sort of following up on that, obviously, there's the BDC, right? Which you sort of launched about a year ago. Help us kind of understand how that sort of fits into that sort of broader data model.
Yeah, for sure. Yeah, Business Data Cloud, just to be clear, it's a SaaS offering, and it's a little bit different than what's on the market. It is not a data lake, okay? You're not extracting... I mean, you can, right? It has components of that, but you're not extracting data into it and then building up a data model and like PAS, okay? Like you would do with other solutions. It sits actually on top of or integrates with partner solutions like Snowflake and Databricks and BigQuery, right? Because customers have invested time in Snowflake, for example, right? SAP and non-SAP data, external data, right? Consolidating that, okay? What Business Data Cloud, it sits on top and it operates on the principle of what are called data products.
We release around, I think, I think we have a target around 500 data products from SAP systems. How do you consistently define a sales order or purchase order, right? Which can have different data models, depending if that's coming from an Ariba system or an S/4 system or whatever. And we expose these as data products, and then customers can also then expose data from, again, all these other data lakes that they've built up as data products, and that's contextually rich data. Now you can put an agent on top of that. If I'm doing, I don't know, planning, for example, for, I don't know, headcount planning, things like this, like hiring planning, you know, I might need to know what are the sales forecasts, right?
How many people do I need to hire? That data might be sitting in a Workday system and an S/4 system, and now we can put Joule on top natively with Business Data Cloud, with all of these data products. It just brings together SAP and non-SAP data, and we monetize it, obviously. In a way that can be easily consumed by agents. It's an important part of it. Yeah.
Just you talked about Joule, and Joule has, like, seen number of customers almost growing 9 x over 2025.
Mm-hmm.
When we look at Joule, can you, like, talk about any of the interesting new use cases that you see emerging? How is for, like, anything on, you know, the productivity, any tangible productivity gains?
Mm.
how you're basically, like, looking at it.
Sure.
In terms of measuring it?
Yeah. Yep. Yeah, I think we discussed that last year, that the adoption curve is pretty unremarkable in the sense it's just a typical adoption curve. You release a product, and then customers look at it, and then they buy it, and then they go live with it. We see this kind of like a hockey stick, basically, right? 9 x the number of customers adopting Joule. No, I think the most exciting thing we're releasing now in Joule are the agents. And let me just digress for a second here. We don't want to get into the number counting game for agents, so we set the bar very, very high for what we call an agent.
You could call any, like, RAG use case, like information search use case, an agent, for example, and I've experienced that in, you know, SharePoint, right? Where it says, "I'm your SharePoint agent," and the only thing it does was summarize the SharePoint page, and I was like, "This is not helpful," right? For us, agents need to have agency, right? They need to plan and iteratively work through several steps and tools for us to call that an agent, okay? Agents work very well right now, like production grade, in these like, narrowly constrained type use cases. Some of the agents we've released, for example, is like accruals, accounting, and finance, right? When do you post your costs and revenues, right?
When you do accruals accounting, that might depend on a PDF document with a policy. It might depend on some email chains, some past decisions, right? It's hard to automate, but an agent can actually iteratively reason through that, these kind of processes, right? It has, like, an explanation of how it reached that decision for the accountant to look in and say, "Okay, yeah, that's right." We think that would save, you know, for a medium-sized company that might spend, I don't know, end of a quarter, 12 hours doing accruals posting, they can get that done in 2 hours, for example, right? Production planning agent, same thing. You're planning your production. What happens is a delivery is late, right?
You don't have the right parts, or you get a big order that comes in that's prioritized, and you have to go back, and it's like it's an optimization problem, right? We have an agent that will iteratively reason through that and optimize your production planning, and you can do this production planning much more frequently now, right? You can do that, like, a couple times a day if you need to, and companies can simply sell more. Right? They can deliver more. Release these kind of agents, but definitely the star of last year was Joule for Consultants. This won a award at the World Economic Forum last month. It's called the MINDS Award for, like, transformative industry use cases.
We did that together with KPMG as a partner/customer on that. You know, one of our top priorities this year is AI-assisted cloud migration and AI-assisted cloud transformation. Getting customers faster with less effort to the cloud, obviously onto standardized SAP landscapes. What Joule for Consultants does, it's grounded in all of the SAP help documentation, internal knowledge base articles. Customers can extend that with their own, like, what are called Business Blueprint in SAP world, right? It leverages a proprietary large language model that we train on SAP code, it's called ABAP LLM. That can, like, explain legacy code, it can write unit tests, it can do a bunch of stuff, right? Long story short, Siemens is a reference customer there.
They said that saves around 10 hours a week per consultant. If a consultant's working a 40-hour week, that's like a 25% productivity boost. I mean, I started in consulting. I think my daily rate was probably $2,000 or $2,500 a day, right? Doing SAP transformations. I mean, that's like a massive, like, boost in terms of productivity.
Great. Let's open it up to the audience. I'm sure there are questions. Who wants to start? At the back, please.
Elena from Millennium. To your last point about Joule for Consultants, do you see that accelerating or lengthening sales cycles for SAP core business? I'm asking because are clients kind of seeing that, "Oh, you know, I can get the 25% faster, so let's do it," or, "Hang on, let's wait. Maybe there is a new iteration coming out that will make it even less costly, even faster or something?
Yeah. Again, the question is, do customers wait for AI to basically automate the entire transformation, right?
Yeah.
To the cloud. It's not a completely unreasonable question, to be honest. I think customers weigh that with the benefits that they get when they get to the cloud and are able to use more AI. I mean, look, and frankly, just to be honest, what Joule for Consultants does now is kind of like just the tip of what's possible, right? Because we also announced our integrated tool chain, right? This is using, like, SAP Signavio, right, with machine learning to map out the process automatically. You know, we're gonna release, like, data deduplication and data harmonization now based on AI, which can be like a huge test automation, right? There's a lot more coming, right, that will help with their cloud migrations...
I don't, I certainly don't see any customers who are waiting, right, under the impression that, "Oh, if I wait a year, then my migration costs will be 50% less?" AI is gonna completely automate it. I think it's more the other way. It already offers quite a good benefit for them to get to the cloud, and they have incentives to get to the cloud because there's more AI that's feasible there.
Nick?
Sean, what are your thoughts on agent commerce-
Yeah.
specifically on how to reach out that?
Yeah.
How do you see that right now?
Yeah, for sure. There's a been a few protocol. I think there's two or three protocols, right? Google just released the UCM, right? There's an Agentic Commerce Protocol. Basically allows agents to shop for you, right, and do financial transactions. No, we are supporting that in the CX portfolio of SAP, and SAP Commerce Cloud. SAP Commerce Cloud, we're actually opening up MCP servers as well, 'cause that creates more value for our customers, right? It allows agents to more easily find products on the websites. Yeah, we certainly aim to support that protocol. I think it's a great opportunity.
Yes, go ahead.
Basically, there's around, let's say, $100 billion of market cap that's built on top of SAP. If you do believe that AI does make the marginal cost of R&D and development closer to zero, and historically, you haven't been able to target these sort of upsells and modules that have been done by third parties. Do you see that as sort of a real opportunity now going forward, or do you think we're still far from that, and a lot of people need to change? An example of that maybe that you guys tried to buy Blacklane last year. Is that something now that's easier to build a similar quality in-house because of AI, or can get there in the next few years, or you think you're still very far from a world like that?
I mean, everything indicates that AI reduces development by, like, 20%-50%, let's say, like, productivity, right? Yeah, to build stuff, okay? To flip it around, there's also that threat, like, will there be new entrants challenging SAP or will... I didn't completely finish your question before. Will customers be more incentivized to build things in-house because it's so easy, right? Using low-code. I think, again, the assets. I mean, again, we also apply that, right, to your point. Now, can we now enter more markets, right? I think, again, building on the capabilities that we have, and again, all this, like, the product standards, the enterprise readiness, the knowledge, and all that kind of stuff is, like, a really important base. You need that to enter these markets.
You know, we are taking some measures to actually do that. This has been disclosed as well. We're doing more FDE, this forward-deployed engineering. This is an initiative from the product and engineering board area, popularized by OpenAI is doing it now and Palantir, right? Basically, you send engineers to customers, right? In 90-day sprints, you figure out real problems they have, then you try and, like, build something, then for us, we wanna build repeatable solutions, right? That's, you know, it's not a new thing, but with low-code, frankly, and AI, it's a lot more feasible to do that in a 90-day sprint now, right?
We can enter all these markets, and we're gonna offer a lot more industry solutions moving ahead to enter those kind of areas where we can certainly build stuff, right? 'Cause, again, customers and others can build stuff more easily, but so can we. Plus, we have these, like, complimentary assets where we have all the customer relationships and knowledge and all this kind of stuff in place. It has the potential to grow our portfolio, at some point.
Yeah, it's the best of breed. Maybe we carry on. Dominic talked about this incremental EUR 1 billion of revenue opportunity for SAP from a number of sectors or segments, and obviously, AI offerings, particularly the cross-sell, upsell of these new, were historically been line of business solutions, is now some of these kind of AI solutions. Maybe talk us through some of the traction you're seeing around this and, you know, to what extent, you know, has this really taken off or what we're seeing from some of the other LLMs now, that the customer's opting to go down the kind of build route there versus, say, buying something from you off the shelf?
Yeah. I think we've disclosed, I mean, the tremendous attach rates, let's say, right?
Yeah.
Two-thirds of the order entry volume, right, was had AI attached, like, 90% of the top biggest deals had AI and BDC in it, right? We see it's a huge part of the sales cycle, right? Honestly, it's an incentive. Now, we support customers to also build, right? That's always been one of the attractive things of SAP systems, is that customers like to get in there and.
Customize.
For better or worse, right, do their customizations, and apparently, they need to run sales orders in their, you know, non-standard way or whatever. They go in and actually build those, so we actually offer tools now, right, like SAP Build, which has, you know, we have client integration, which is one of these low-code apps, right, that they can actually use to build and extend around SAP systems. Of course, we give them that option, and that's always been a big part of our portfolio. You know, it still needs to be based, I think, on you need the standard core.
Right. maybe just moving on, Oh, sorry. Yeah, go ahead, please.
Too much costs. Yeah, the providers and some of the startups kind of running on cash flow, actually, KPI is not necessarily-
when you're offering a similar product, what does it do to pricing? Does the pricing compensation happen without copying? Thank you.
Good, good point. Two things. One, for the embedded AI that we offer, okay, not the low-code, but the embedded AI, right? The AI, and we sell these by AI units. These are not tokens, to be clear, right? Tokens are part of that margin, but the business value is also part of that margin. When we release... Again, everyone's always pitching use case ideas to me, and I've seen, like, I feel like I've seen all of them now, right? We put those in our products, and we can build the embedded stuff, right, once, which includes, you know, testing it, running an ethics review, benchmarking, right, building the UI integration, all this kind of stuff, right? We'll build that once, but then we can release that to 5,000 customers, and that's the margin that we take on top.
The token cost, the model cost is just part of that. Again, we have this strategy where we partnered super broadly. We have access to all the Frontier models. We have partnerships with Mistral and Cohere and all these companies, right? We have some arbitrage, right? We can move to the cheapest, best performing model as we want under the hood, okay? That, in that space, we have the margin, right? There's a benefit for the customer there. When it comes to pure low-code, the offerings that we have are, let me put it this way, a little more broader than just the low-code part of it, right?
If you're looking at SAP Build, right, there's the low-code component of that, which they may undercut us or try to, like, you know, push the price down. There's all the other components if you're using SAP Build, right? Like, you need to ramp up a database, and you need to build a CI/CD pipeline and all this kind of components around it that, I think is where we make our margin and can, like, sell above and beyond what they offer.
Yeah, at the back there.
Talk about how you guys are protecting the sort of cross-sell opportunity, assuming that customers might code themselves on the edge, basically, on edge opportunities?
Mm-hmm.
Touch on that a little bit, how that affects you guys.
SAP systems are built to be customized and extended, right? We support that. I think where we have a strength, again, with the broad portfolio that we have and things like Business Data Cloud, is a lot of the, like, value-added use cases are not just point solutions anymore. It's solutions that, or extensions, right, that touch several, like, business processes and business applications. The planning scenario, for example, that I mentioned before, right? You might need data from logistics system and finance and HR, right? We can bring together that data in Business Data Cloud, right? We monetize that. You can do a low-code and extend an agent on these kind of solutions. I think that just strengthens the whole suite story.
Then, in terms of customers building on the edge, I mean, that's something we historically support anyway. That's one of the attractive things about SAP systems, so yeah.
I guess one of the, part of what's driving a lot of the consternation, when it comes to SaaS versus AI, it appears to be that a lot of the innovation seems to be happening at the labs startup level. So even things like, you know, OpenClaw, for example, you know, why did OpenClaw even exist? It should have been built by Microsoft, if they already had Copilot, and no one uses Copilot but OpenClaw. I wonder from your perspective, as you know, given your seat, do you feel like you have the right team in place to be able to remain relevant and be at the Frontier of all these developments?
At the end of the day, switching costs are quite low, and so if someone can use a cloud code versus using SAP's, you know, Agent Builder, it could use less than SaaS. Obviously, there are clients and data sets that make it more difficult for them. Surely, not a good idea to force your customers to use a product. I'm just trying to think about why the cadence of innovation is being so slow with the existing AI software, and yet all these software companies seem to be very much behind the curve from a new product adoption perspective and all the headlines and graphs.
Yeah. Yeah. Did anybody install OpenClaw on your computer?
I know.
Oh, man, you're brave. It's a security nightmare. It's impressive, though, what's possible with it. Yeah, I mean, I disagree that the switching costs are that low, to be honest. Again, I do think there are benefits if, like, if you're building SAP extensions, like, a lot of our big customers are SAP shops, and they're used to working in the system, and they work there because of, again, the enterprise readiness that we have around it, like log, auditing and all this kind of stuff that's needed. I'd challenge a little bit about the innovation piece. Again, like this RPT-1 model that we released, this tabular model, I mean, that is like, again, spotlight paper at the top AI conference in the world, like, huge value. This, company, I think they're called...
What are they called? Frontier, I think, just came out of stealth mode, right? Building tabular AI models with a valuation of over $1 billion. I mean, we have the same thing, basically. I mean, to be honest, even better, I would say, right? No, I mean, I think, again, I would challenge that a little bit. I do think there is some innovation coming out of the big companies. Again, we're mostly focused on making things enterprise-ready and productizing. Also, again, a big part of our strategy is partnering very broadly with these companies. Look, we have partnerships with OpenAI, right? We made this announcement. We're hosting OpenAI models on Sovereign Cloud in Germany, for example, right? We're investors in Anthropic. We just participated. We're early investors.
We just participated in the last round as well, right? We do stay close to them and, you know, let them experiment, okay, and fail and be successful in certain areas, and then when they're successful there, that's where we partner.
In a world where, I guess, ERP transitions happen in short time, maybe months rather than years, to your point, actually.
Mm.
-managed by AI completely, like, what's your outlook for the systems integrator ecosystem that depends on?
I was always stuck with the statistic I'd seen. Every $1 sold in SAP software generates, like, $6-$10 in the ecosystem, right? For migrations, maintenance, all this kind of stuff. No, I mean, honestly, to their credit, they are also actively disrupting themselves, right? Joule for Consultants is used by all the big systems integrators, right? They dive into the customer's ask for it. You know, their perspective is, they can do more projects in a shorter amount of time, right? There's no shortage of projects and customers that need to be migrated. They have trouble finding talent, right? They're pivoting their business models very quickly.
to their credit, I mean, they've really been also actively embracing this and disrupting themselves, and I think seeing a lot of growth and success as a result.
Great. Well, I think we're on time, so thank you very much, Sean, for the great insights, and thanks, everyone, for joining.
Just give a word of the session there.