Hi, everyone. Thank you all for joining our investor event today. I would first like to go over a few housekeeping items. Today's session is being recorded and will be webcast live. I would also like to point you to our Safe Harbor Statement, and to remind you that today's discussion may contain forward-looking statements. Actual results may differ materially from these statements as a result of various risk factors, including those found in our SEC filings. We may disclose information related to development and plans for future products, features, or enhancements, which are subject to change at our discretion without notice. All statements are made only as of today, and UiPath undertakes no obligation to update any forward-looking statements and makes no assurances and assumes no responsibility to introduce future products, features, or enhancements described today.
Additionally, we would like to note that we are in an earnings quiet period, and as such, we will not be taking any financial-related questions. In terms of today's agenda, we'll begin with our CEO, Daniel Dines, who will lay out our, his vision for the company. Then Graham Sheldon, UiPath's Chief Product Officer, will give a review of recent product updates and demos. This will be followed by a customer panel hosted by Ben Feichtner, who leads our go-to-market efforts in Americas with IAG and Dentsu, who will discuss how they are harnessing the power of AI and automation together with a brief Q&A session for those attending in person. Following the customer panel, we will take additional product-related questions. Afterward, we will move to another room for a cocktail reception, where you will have the opportunity to interact with additional management, with additional members of our management team.
With that, I'd like to introduce our CEO, Daniel Dines.
Hello, everyone. Thank you so much for being here with us. I know many of you for many years so far, and you've been with us for good days, sometimes for bad days. But today, tomorrow, and day after tomorrow is gonna be good days, guys. I promise you. It's basically we are announcing the second act of UiPath, and I want to give you a quick glimpse into what's coming and why we believe it's our second act. In the past 10 years, we've been building our platform, business automation platform, to go after rule-based processes. As long as in a process, the information was structured and the steps that a person should follow were governed by rules, easily to express, then we could go and automate the process. I think we did quite a good job.
We started from like a small apartment, our equivalent of a garage, and we emerge, you know, like a billion-dollar company going into, I don't know, more than thirty countries with offices and everything, going public on New York Stock Exchange. So it was really a good ride, and we made a platform that was capable of delivering really reliable automations, but it was limited, and we always knew it was limited. Many business processes are not really comprised only of rule-based or structured information. There is a lot of unstructured, dynamic steps in a process that we couldn't automate. So our strategy was always to kind of untangle what is structured, what is unstructured, automate what is structured, and let people handle the unstructured part, and it works well, but I think we left out a lot of processes.
Because think about our customers, if we get into a process where maybe 20% is structured and 80% is unstructured, there is little point to go after the 20%. It's not, it's not enough value that you can get there. But now we are in a fantastic moment when the technology to deal with the 80% unstructured is emerging, and we can tap into it. For the past 18 months or so, I think us, many other companies, many of our customers are trying to figure out, how can we get the power of generative AI and deliver it into an enterprise automation scenario, deliver it in a reliable way? Because in a way, this is the crux of the issue. This technology is really powerful, but it's unreliable, and you will need to find a way to make it reliable.
I think a lot of progress happened in the industry, and this concept of agents are all the rage now. For us, it's a really big moment because I will make a point that UiPath is a company that is really well-positioned to deliver on agents, on agentic automation, to deliver on a technology that can grow also to address the unstructured part of a process. Because in a way, it's a natural progression of RPA. RPA, it's a natural progression from rule-based automation into a more dynamic, unstructured type of automation. We are going to introduce the notion of agents. An agent is basically a combination between an instruction set to an LLM and a set of tools that the agent can call dynamically. These tools are, in fact, robots in our platform.
So I want to give you a very quick example that we can all relate to. If I want to build a travel agent, right? I have to. I will give a tool to the agent to retrieve the company, my company travel policy. That can be updated from time to time, so it's dynamic in nature. So agent will retrieve the company policy, and then I will give agent a tool to retrieve my information from Workday. What's my position in the company? So the agent can understand what kind of allowance do I have, at what kind of flight tickets can offer me, and hotel rates and everything. And then I give the agent tools to find, to search for flights, to search for hotels, and then I give the agents tools to book a flight and book a hotel.
Different airlines, different type of websites, aggregators, everything. Now, the key is that this agent is capable of dynamically choosing the tools based on the input. So if I'm going and I say, "I want to book me... I'm gonna go to-- I'm in London, I'm going to go to New York next week, and I want to book me a flight and a hotel and return in two days." All in natural language, agent is capable of understanding intent and choose different tools that has at its disposition. The tools are reliable. This is a very key aspect. You can rely on the tools, and tools abstract away the information in a way to the agent. Because you don't want to give an LLM agent your credentials to work with, because you don't know what's gonna do there.
But if you give a tool that is very precise and reliable in nature, the agent can use the tool, and you have the certainty, the information that is going there. It's only the information needed for the particular task you go there. Now, we at UiPath has created these integrations with different platforms. This is our specialty. We are a Switzerland of business applications. So we have all these reliable tools that can help an agent thrive and really go and fulfill the goal that, you know, someone is given the agent. And we have all the governance and security to run the agents as much as we run our own robots. And we are creating, we are letting our people, using our low-code platform, to create these agents.
We tap into our big community of people, like we have three million people trained in our platform, to also build these agents. They are a natural part of our platform. Because in a way, it's as in Workday, you can have low-skilled individuals, you can have high-skilled individuals, but they are managed by the same platform. It's kind of natural to have our platform managing robots. In a way, you can see them as low-skilled employees and agents, which are more high-skilled employees, and the agents will take advantage of the low-skilled robots in order to perform their high-level job. Everything is surrounded by our orchestration engine, because we have built orchestration since the beginning. There is no way you can deliver robots in a reliable way without having orchestration between robots and humans.
And now we are extending our orchestration to have agents talking to humans, talking to robots, and everything in this workflow, enterprise workflow, that is really reliable. Look, in the past, like, six weeks, I traveled the world. I was... A week ago, I was in Singapore and then Dubai, London, and then here in Vegas, and prior to Tokyo. So I talked to a lot of our customers. It resonates so much. This agentic automation thing opens the discussion. They are really open to all these new use cases. They are open to test, they are open to co-innovate with us.
It really makes sense when I mention to them, "Guys, in order to deliver AI in an agentic shape, autonomously, you need process orchestration, and you need to have robots talking to agents and to people, and ground the agents using robot in the enterprise data." It all makes sense. I've seen CIOs saying, "Daniel, there are many companies that talk about AI and automation, but now I see UiPath in a different light. Process orchestration plus AI plus automation is the key to deliver." Again, we are the Switzerland of platforms. We connect agnostically to all the business applications. We have connectors, we have robots that can go to all of these platforms equally. Salesforce can build agents, but they will build agents that live and thrive into Salesforce. We can even connect to their agents like we use APIs to different platforms.
But think about most SMEs don't use one single business platform. They use multiple business platform. So this is why an agent that mimics an SME, that use multiple business platforms, will need to have robots that are capable of connecting to all of these platforms. This is where our unique advantage is. And I think Graham will show you practically what it means, where we are from a product standpoint. But I want to finish saying it really brings a lot of energy within UiPath. Everybody right now has a singular focus: let's bring agentic automation to the market. It's really our biggest opportunity, and it's the beginning of our second act. Graham?
Thanks, Daniel. So for those of you who I haven't met, I'm Graham Sheldon, the Chief Product Officer. So the product vision, strategy, and roadmap are things that I get to think about all the time, and I get to talk with Daniel about those big ideas and how to make them real. So, mostly what I'm going to do is show you what we mean by agents and robots working together. So what we are investing in, as a company, to make this agentic automation strategy real, are four key things. The first is bringing to life enterprise agents. There are a lot of agents out there, as Daniel described, but for ones that you're going to actually deploy for your mission-critical business processes, for these end-to-end workflows, you need to be able to trust them.
There are things that we are going to do specifically to bring those agents to life in a way that you can guarantee that the decisions that they're going to make are ones that you can trust, and that they make these decisions independently so that you can remove as much possible from the workflows that the humans are currently still doing today. That's something that is a big advancement for generative AI, and it's kind of like... I think it may actually be helpful if I start by describing why this is such a big leap.
I mean, for those of us who are in it, I think it's useful to level set that, like, there was massive gains in productivity when the PC came out, and we digitized a lot of the physical work that happened. There was another inflection point and a disruptive point when the internet came around, and you had the democratization of things, and then, you know, with mobile productivity, mobile applications, putting all of that in your pocket, just at your fingertips. I think we're all going to look back at this moment and think about all the work that has been automated in the past as being only half of the problem. It's the left brain stuff, so think about your work, like the things that you do that are based on rules.
If I get this kind of a case, I do that with it. If I get this kind of an invoice and it matches, great, I'll send it straight through to process. The other half of that problem, though, is what happens when you don't have a business rule to describe it, where you can't ask a developer to sit down and code all of the different variations and all of the different error cases that you might run into. So you need to have the right brain stuff, the more creative things, the more adaptive, the more flexible.
What UiPath is going to be really great at is combining the left brain stuff that you're doing with rules that we will continue to invest in best-in-class automation for, with the best of what generative AI can now do with large language models and large action models that can help you anticipate what's going on and apply these things in the more non-deterministic cases, where you're applying a policy, like Daniel described, for travel, or a policy for taking on, you know, a mortgage statements, and you're trying to approve a loan. Those are cases that apply a little more judgment. There are a lot of these cases where today, there's many, many use cases. I'll talk about one of them in a little bit.
But many of those use cases, you know, are things that today you have to rely on people to do. You have outsourcers who, you know, you are not willing to go pay a full-time employee for, but you think that you can't just write business rules to do it. Those are the kinds of things where agentic orchestration comes in play, where the best way to solve those problems, again, end-to-end, will be to put the robots to work on the tasks that are deterministic, that are based on rules. Put agents to work on the things that are based on the goals that they're trying to seek, and obviously, people for the most critical end decisions, so we're not going to completely remove people from this, but they're going to be put to use for the right problem.
We're going to continue to invest in our core automation platform because those agents, this is a really critical part of the strategy. Those agents are only as good as the tools that they have. For many of you who played around with ChatGPT, especially early on, you know that, you know, you get somewhere, and then you've got to go copy and paste that data into some other system, or you got to go copy and paste that data back into ChatGPT to get anything really done. That's not useful to have yet another tool in this panoply of you know, in an enterprise ecosystem. You want that stuff integrated, and so that integrated value and what the automations can bring is a way to do that and a way to do that securely.
So those automations are built against your ERP system, your CRM system, your HRM system in a way that you trust it to do the right things. So if you give the agents those tools, they can be the best agents. And because, again, we are the Switzerland of those, we've invested a lot, both over the last years and will continue to make sure we have the best automations, which become the best tools for those agents. And then, of course, we continue to work on making sure that we meet all of the highest level of rigor in terms of regulated industries and customers who need to meet their compliance, very high level of compliance, and governance with our cloud. So a lot of this innovation is obviously happening there, and that's where we're seeing a lot of growth.
So it's these, the combination of these things together that will help us be in the best position to help our customers build agents that they can trust to do a lot more work than was ever before possible. Our first best agent that we created last year and announced at this conference was Autopilot. And Autopilot took flight earlier in June, and it's seen a lot of great adoption and success. I was just talking today with a whole bunch of customers who came up to me and said, "Oh," and were telling me about use cases that they were using it for. Cisco is one here where for testing it's resulted in a huge reduction in manual effort and yearly savings.
I had another customer come up to me today and say how it was helping them with filling out employee self-evaluations. You know, who loves filling out employee self-evaluations? It's better if you can, like, just take all of the content for the stuff that you've been working on and have Autopilot help you put it into the right format, so you can put it into Workday, instead of sitting in front of Workday, filling this stuff out. These are just a couple of the use cases that were really. Flo is here from Dentsu in the back. Hi, Flo. Nice to see you. They've done some really exciting stuff, applying some of the HR policies, and helping exactly in that scenario I just mentioned.
WEX is doing some real interesting work, where it's taking different sources from Google Docs and from Google Calendar and some of the transcriptions and things, and helping people prepare for their meetings. I know every time I go get to meet with Daniel, he asks me a hundred questions, so it's great to know beforehand, you know, what it was that we were talking about. So having an in WEX's case, they're using Autopilot as a way, an agent for, to help them prepare for those executive meetings, to prepare the action items and key points. And then the last one I wanna show you is one that's kinda interesting, not only because it's saving time and employees' experience is getting better, but also because it's saving lives.
So CSL is a company that, Behring in particular, is one that is, they are, they're working on, taking in donors of plasma. And so we've been working with them to develop Autopilot as a way to make it easy for the physicians to help triage the patients who want to be able to donate and are using a cross-section. The folks, I mean, it's amazing. They have to sit down with a huge, long document, like hundreds of pages, that have all the medications and types of complications and things like that, that govern whether or not they're allowed to take your donation.
And that's critical because when they deliver those to their customers out in the field, if they get any of that wrong, someone transcribes it incorrectly, they have to get rid of that plasma. So that's a person's life that's now at stake by making sure that they don't have to do that. And I actually wanna show that to you directly. So with Autopilot, it combines basically the best of what generative AI can do in terms of understanding natural language, plus what UiPath is doing in AI, called specialized AI models.
So in this particular case, if you're sitting down with a donor and who has psoriasis and has also taken Accutane within the last three months, you know, the doctor might remember that, but there's obviously a limit to what they can remember. So this is going and checking against all of the documentation, those sort of standard operating procedures, to see whether or not that's someone that they can take, and indeed comes back not only with an answer, but with the documentation behind it. So you can take a look at that, and that the doctor can actually refer to the source data, the citation from that source data, to be really sure that they're about to make that critical decision. It's a good example of where you are developing that trust.
When I'm ready to, I can then go and onboard this donor. This is where we're using specialized AI. So there's a specific model for ID cards that Autopilot has, that it's going to make use of to take out the information from this driver's license. And for those of you who are on the earnings calls, I've shown this particular one before, where in this case, if I was entering this all by myself, then I'd have to be waiting, like the patient is sitting there, you know, wondering what's happening. It would take me, you know, 20 minutes to enter all this information in. Instead, this is going to. Autopilot pre-fills out an automation that's going to drive the UI. So because Autopilot is built on the UiPath Automation platform, it can now drive this whole experience.
So you can see that it's doing this while we're waiting, and it's just doing it all for me. Because I'm in the loop, because I'm watching it happen, if there are any mistakes that get made, I can catch those mistakes as we go along. But it's saving me a ton of time, saving me some of those clerical errors that would ultimately end up in maybe some of those that plasma being discarded later on down the line. Because many of the doctors do this very often, there's a component built into Autopilot, and this is one of the bigger differentiators for Autopilot, that remembers how I did it before. So that memory component allows Autopilot to learn what the next step is.
As you probably know, doctors don't always remember to put their notes in, or they wait till the end because it takes a long time. In this case, because Autopilot has the whole chat history, it summarized all of that. It suggested that I enter it in, and voila, it's entering all of those notes in for me. I didn't tell it to do that. It remembered and prompted me proactively to do that work. So we're driving the business process and some of those best practices by using Autopilot and its capabilities. Autopilot also has this ability to understand unstructured data. As you notice, this medication list has a bunch of, like, handwritten stuff in it. So it's not just structured data, it's a little bit unstructured, too.
This is critical information to grab from the patient, because these medications have cross effects in them that might affect their eligibility, and so I need all of this critical data in a form that we. Last year, Clipboard AI is something that we announced and won the TIME Innovators award, and as part of that, what Clipboard AI can do, not only is to understand the data that I'm putting in here in structured data, but it also understands screens, so in this particular case, I'm gonna use Clipboard AI, and Clipboard AI is going to look at the form on the left-hand side. I have not built an automation for the thing on the left-hand side. I could have, but I didn't.
What Clipboard AI lets me do is allows Autopilot to examine what that form actually has, understands what it's asking for, and translates the data I have about the medications into that form. So now I'm gonna be very precisely be able to fill this out. So when I hit paste to put the, that data in, it'll allow me to do that. And you obviously, at least in my household, we have lots of medications 'cause I got three kids, and if I'm gonna put them all in there, then we, you know, you tend to have a lot more than even just this. So this is a massive time savings for someone to use Clipboard AI within Autopilot to be able to paste this into the medication form.
There's a lot of these types of processes that you can obviously imagine that this would be useful for, but we're not stopping there. This is a sort of reactive digital assistant kind of experience, but we're talking today about agentic AI, independent, autonomous agents working on my behalf. I just did a whole bunch of work there, but wouldn't it be better if I didn't have to do that work in the first place? Maybe it would be better if we could actually build an agent to help me process the easy cases, so that I could focus on the ones that are harder, the ones where real judgment is involved. In this case, we're gonna go further with Autopilot to ask it to actually generate agents for me.
In this case, so it's going to use what it's done in the past, my, the memory that it's got of what I've been doing, and create a new agent that will go through my donors. So you can see it proactively came up with the donors that are waiting for my action, and it's going to create a flow for me that this agent is going to follow. So I've confidence that it has the right flow, that it'll use this basic loop to go through my donors that are open. And then, it lets me basically just use the automations that I have to approve the ones that are good, reject the ones that aren't, and leave the ones for me, waiting with a recommendation that it's not clear about.
And then, the last step in this is I can actually go ask for Autopilot to schedule this for me. So just imagine, you're a, you're walking in, your first cup of coffee. Autopilot is already completed for you most of this work, and you're left with just the ones that you need to pay the most attention to. The rest of this demo, I'm not gonna be able to show you in the interest of time, but as you can see, it's spawned a new agent on the side that's gonna work for me. I can multitask. I can keep talking to Autopilot if I want to, but it's going to see those three agents, and it's gonna start processing them on my behalf.
All right, the last thing, and Ashim asked me to try to get this short, and I'm going to fail at this. But the last thing I want to give you is a use case for invoice processing, and this is where we're gonna create a new agent. So this is where customers, in the past, for those of you who don't know about invoice processing, robots have been super helpful. They've taken a lot of the pain out of doing the two-way matching and entering the data into the ERP system. But what happens if there's a mismatch? There's still a ton of human work that has to happen. Now, I've got... If there's a mismatch, now I've got to go back into the parts database. I gotta go see if this is a substitute for that part.
If they returned it, maybe they did it within the thirty-day timeframe that I gave them, or if it's over sixty days, then I'm gonna have to charge them for a replacement. There's many different policies at play here that a human has to do. Then you go back to the supplier, and you're back and forth with the supplier. "Hey, you charged me too much. You've added an extra zero. Can you revert these things?" There's a ton more work that still is manual in this process. That's where we think agents can really play a big part. Applying those policies, communicating with the supplier, and making some of these decisions automatically so that the humans don't have to. The humans are just there to do the final approvals and to check that the right decisions are being made.
So here's how this is going to work. This is our new experience for building agents in Autopilot, and we'll obviously build. This is a catalog of ones that we think are great, that we'll have out of the box. This one happens to be a dispute investigation agent, which I can open up in the Agent Builder. But I've already done a little bit of work here and specified some things. So this is the Agent Builder experience. In the Agent Builder experience, there's some critical things that I can do to build, test, and deploy these agents at scale and in my mission-critical things. As I define it, there's some basic stuff I need to do, like a prompt. The prompt defines what the role is that this agent will serve and the goal that it is seeking to accomplish.
The second thing, and this is really critical, are the tools. And I told you before that the tools really matter. The agents are only as good as the tools that are in there. In this case, it has access to SAP, and it has a web reader to research about the parts that the supplier has given me. But those tools are not limited to those. It's every single automation that this customer has access to, every activity, every API, every agent, every process, and other agents, ours, and other agents from other ecosystems, that are now available as tools for this agent.
That's super powerful, because now I can say, "These are the specific tools I want you to use," and I can say, "This is how I want you to use them, and this, this is how I don't want you to use them," so that you're more likely to get an accurate response and one that you can rely on. Importantly, there's context. So in this case, there's a policy document that we've indexed and used the enterprise context service form. And then, crucially, and this is another unique part of the UiPath approach to agents, I'm defining an escalation. If the agent needs help, and it can't find the supplier's email, it needs that information to kick off the next step.
If it can't find that, we've built this in so that it can go to Action Center and ask someone specifically for that piece of information, so it stays on the guardrails. All right, now we're gonna give this a run. I've specified some input data, a couple of things that don't match, and now it's gonna go try to figure this out. What the agent is doing right now, as I'm testing it, is it's coming up with a first, a dynamic plan. It happened to have picked a couple of tasks to do with the web reader to go get information about the mismatched, matched parts first. Then, it's going to go to the context, the policy document, and it's going to figure out which policies apply and which ones don't apply. The last thing it's gonna go do is go into SAP.
In this particular case, we did this as a UI automation, but it could be an API automation. And again, this is unique to the UiPath platform, that it's going to have access to these particular tools. And last but not least, it ran into a problem. It didn't find that email address for the supplier. So what's it gonna do? It's gonna escalate. Now, the agent is paused. It's waiting for critical input before it moves forward. It's gonna go to Action Center, and it's gonna assign this to me. We've done this in Action Center, but these will show up in Teams or Slack or wherever or in email, and here I'm gonna give it the reason for, "Here's the thing you should do," and the reason. And down at the bottom, it says, "Add to agent memory." It went really fast.
Sorry you didn't get to see that, but that memory thing is what allows us to learn. It's that memory part, when I give it that explanation, the next time it sees a case like this, it knows what to do. So now that we've done that, we really want to figure out... That, that's great that it worked in one case. We want to figure out if it works in a lot of other cases. To have real confidence in this, I need to be able to evaluate this on a whole bunch of different cases and make sure that it's going to meet my criteria. I want, for this kind of thing, I want it to meet, you know, 90, 90% score.
So over time, I can see we will run these evaluation sets and give people the ability to see how well it's doing, and then it can also capture all of the examples that it has seen so that I can actually go back and inspect the kinds of decisions that it has made and tweak it as I need to. These are critical capabilities in order for you to build these things that actually will work reliably and with high confidence. Last but not least, we're gonna go create an experience. So there's two ways to do that. One, I can create an app, so that's that digital chatbot kind of experience, or I can include this agent in one of my existing workflows.
Because we have these structured inputs and structured outputs from the agent, I can include it into these workflows. This is my invoice dispute agent workflow, where I've included this particular agent in there. Any existing workflow, any existing automation, you can now extend with these agents to take care of those particular decisions that need to be made. All right. I think I'm gonna end it there. We've got a lot of this stuff is new, the Agent Builder Experience, the Agent Catalog, the ability to build apps that use the agents, the agent service that does the planning, the context, and learning. And we're obviously quite, we're here to, during this conference to figure out which use cases this is gonna apply best to.
We've got a lot of good examples, but the reason that we win on this is, and the reason that we think that we're in a unique position to drive a lot of this AI transformation through agents and agentic automation, is because we have the best tools. The automation platform is best in class. The second big piece is that we're doing this orchestration across robots and agents. We are Switzerland, so Daniel said it. We are ecosystem agnostic. We integrate with everybody. We do the long tail of integration, and then it's on top of a system that is secure and compliant and enterprise grade, so that the accuracy and reliability that you get out of it is top-notch, and that you can run your most critical business processes on it.
With that, I'm gonna turn things over, I think, to Ben. Hopefully, that was educational. I'm sure there's a lot of questions. I went through a ton of material there. Happy to stick around a little later and answer any questions you might have. Thank you.
... You guys wanna come up? All righty. So we have two of our esteemed customers here today, and clients. So we'll have them introduce themselves quickly. So we're gonna go through this. We're gonna answer. I'm gonna ask them a couple questions in a fireside chat format, and then we'll hopefully have time for one or two questions from the analysts, and then we'll hand it over to management Q&A. You guys wanna take a seat? Thank you again for joining us. So I'm sure many of the folks in this room have heard about both of your companies, but I would love maybe to start with a brief introduction. We know we have Dentsu, the global advertising powerhouse, and then we have IAG, one of the world's leading airline companies, and, you know, parent of companies like British Airways.
Maybe why don't you just give us a brief introduction, and then share a little bit on the industry dynamics that's going on for each of you right now.
Sure.
Alex, go.
Let me start. My name is Alex Bédier, CEO of Dentsu. As you know, Dentsu is operating in the media and creative business. It's a Japanese company, headquartered in Tokyo, which has more than 100 years tenure. We've always been working around innovation and trying, I would say, to solve challenges for our clients in a sustainable, I would say, and ethical way. More recently, we've really reorganized our overall promise to become fully integrated across our customers, and by taking the power of all of our agencies, we are now operating as a single, I would say, group across the world, in 20 countries, 71,000 employees. It's a people business that's trying to solve, I would say, the media challenge, the experience transformation through creativity.
An interesting challenge. And as you know, in terms of industry, it's a very challenged industry, I would say. Fast-moving, very, I would say, disturbed by technology, generally speaking, so, we have to be at the forefront of every single challenge, basically.
Awesome. Thank you. Luis?
Yes, hi, everyone. My name is Luis. I look after intelligent automation for IAG. IAG is a group of airlines. We're better known by our brands, British Airways, Iberia, Aer Lingus, and a few others. And we operate in an incredibly competitive space, as you know, so innovation for us is incredibly important, whether that is investing on a panel of startups every year to develop new technologies for us or finding sustainable fuel. We're the first group that commits to net zero, which is a really tricky thing to do for an airline. And obviously, automation and AI plays a huge role for us, so we're investing quite heavily on that to make a difference for our customers going forward.
That's fantastic, and two very distinctly different businesses in a lot of ways. Maybe we'll start with this, Alex, 'cause I know you guys have been on the journey for a few years, but as you've started with the core platform, you've moved into intelligent document processing and Automation Cloud. I'd love to hear a little bit about some of the use cases you guys are solving.
Yeah, sure. You're right. We've been partnering together since 2017. We've been through that whole journey of taking the benefit of the cloud. 700, I would say, automation live today, so I won't go through all of them, obviously. Maybe just, you know, it's 800K of recurring hours, say, for our business so far. Maybe one or two interesting use cases, one which is more related to our client work, where we've by leveraging both, I would say, generative AI capabilities, so context grounding, document, I would say, understanding, and automation, we've been able to automate a lot of manual processes around, I would say, e-commerce, KPI calculation, media optimization for a dedicated client. That's one example.
A second use case, which is a POC still, that we mentioned in the introduction, is: how do we make the life of our employee? We are a people business, so for us, it's very important, more easy when it comes to year-end processes, and again, through context grounding, document analysis, Teams, and that's important, it's an interesting point, build a process to augment the process, helping, I would say, our people to gather information from our policy, from their goals, from how they've been performing. So I like to name that new approach... you talk about Switzerland-
Yeah.
But for me, it's the Swiss Army knife.
Yeah.
But you can, I would say, put that in the hands of end users or people in departments to really transform at the department level, those automation.
That was awesome.
Mm.
Thank you. And, Luis, I know similar to Alex, you've been on the platform for a minute, if a bit longer than a minute, I suppose. But when you look at it, too, as you've expanded into, I believe, Document Understanding-
Right.
Communications Mining, and even our process and discovery suite, you've done some cool stuff, so I'd love for you to share some of the end-to-end use cases.
Yes. Yes, of course. So airlines, as you know, was one of the businesses that went into complete shutdown during the pandemic. So for us, coming back to pre-pandemic volumes, automation, UiPath, in particular, was more of a survival tool more than anything. It was really important to get some processes automated. But we're automating processes not just on you know, your traditional finance, HR, procurement, but very deep niche aviation processes that have to do with how crews are allocated to aircraft, how engines are repaired. You know, when... And I'm sure everybody has experienced this: when you're checking in, it takes a long time. You don't know what the heck people are typing in, but they're typing in for a long time.
So we're working quite heavily on that to try to make that process easier for, for our staff as well. Yeah.
That thought crossed my mind on my flight here on the way in yesterday.
Yeah.
So I appreciate it, but it wasn't one of your airlines.
No.
I will say that. Yours were automated. Well, that's great, and I think you know, Luis, maybe I'll come back to you. We hinted on this a little bit, right? You guys are both doing things with automation and generative AI. I'm sure if you're in the audience right now, you're wondering, does this complement or does a generative AI potentially replace automation? I'd love your take and maybe some of the use cases you're working through there.
No, absolutely. We take a very special view of this. We don't really see automation and AI as being two different things. We-
Mm-hmm
We do see the approach we take is we look at an operation, we look at a specific business process, and we break it down. And we realize that this particular business process has, I don't know, 15 different tasks. And then, what we can do with UiPath is to say, "Task number one is about extracting stuff from a document. Oh, I have a tool for that." And step number two is processing an email. Step number three is making sense of a particular piece of text, et cetera. So the way we look at it is rather than thinking about it as two different technologies, it just becomes tools in our arsenal that we can use to automate processes.
We're doing it today for engine maintenance, where we started with LLMs to digest very, very complex PDF documents, manual to repair an engine. So I don't know if any of you have seen a manual to repair a Rolls-Royce engine, but you're talking about, you know, document this thick. LLMs are incredibly powerful to summarize and answer questions quite accurately. But what we're doing now is, it's okay for an engineer to find the answer much quicker than before, but what if the answer is, "You're gonna need, you know, this rubber gasket, which is part number twenty-three, and that and that?" Wouldn't it be great if we can allow that engineer to say, to allow the technology to say, "Would you like to order that part?
“We have 10 in the warehouse,” so that, to the engineer, is just part of the same process, even though to us in this room, you know, you're talking about RPA now helping the AI. What we're trying to do is just to use those two technologies to complement each other and automate end-to-end.
I think that's a very, very cool use case that clearly defines it for me of the differences between, you know, how you can leverage the power of an LLM and then also automate that action to SAP or whatever your ordering system may be.
Right.
That's great, and Alex, how about you? How do you look at these things as complementary or-
Well, I look at really definitely complementary, and I think we shouldn't underestimate the complexity of this new world, so I'll take my personal example. In Dentsu, we are running today 59 different LLMs, you know, from the very small ones, and there's not one single answer to any, I would say, single problem. Each one has its advantages, each one has its complexity, each one has its cost also. I think that's an important element. At the end, each most of the time where we want to use that, it's like you, Luis, to deliver an outcome and a business process and trying to augment that as much as we can, so for me, yes, it's totally complementary. I'll take an example maybe to illustrate that.
We've been working hard with our legal teams to see how we can really optimize all the journey around contract creation. Again, we are a professional services firm generating tons of contracts. It's a heavy workload. It's painful for the clients, painful for us. So here we are using a combination of a niche LLM technology, which is really, really powerful for contracts and safe, I would say. That always gives you the same result when you're asking a question, which is, for a legal person, very important, and automation to bridge a gap. So automation typically will help us analyze and extract data and tag metadata into the contracts, query that on the LLM to ultimately automate redlining, automate a lot of stuff.
The two are working hand in hand, and I would say one is helping the other to achieve, again, an outcome, which is optimizing our journey, I would say, to contract simplification. And I've got many other ones, but, yeah, that's an interesting one, I guess.
No, I think those are great use cases, and awesome just to highlight kinda Graham's point on bring your own, bring the third parties, use our specialized models. It's very cool that you guys have thought through some of these use cases already, and I guess that ties me to my last question before we open up to the audience here, so when you look at agentic automation and kind of the future long term, Alex, maybe you wanna start with how you guys are thinking about this now, but also in the foreseeable future with everything going on in the environment.
Yeah. So of course, there's a lot has to be done, but it's a great opportunity, I guess. If I talk about one of my core businesses, which is the media supply chain, basically from buying, to planning, to scheduling, not only this is a very complex supply chain, meaning that in each country is very different. We've got many, I would say, channels, you know, social networks appearing every day, digital, TV, media. And to add on that, clients want a bespoke solution almost every time. So I really see this, so obviously, we do automation already, in a lot of areas, but I see agentic really as the secret weapon to help, first of all, going deeper into those automation scenarios.
Help us, I would say, really manage all of the edge cases, and there are many, that cannot be, I would say, rule-based, to your point-
Yep
... helping that level. And when we have to make changes because a client decides to operate differently, to do it very fast without rebuilding absolutely everything. And I'm really positive that, you know, agentic AI and automation together can really crack that use case, and, I'm pretty sure we're gonna do some very interesting, work around that.
That's awesome. Luis?
It's incredibly important for us, and that's reflected by the fact that, in our group, the head of automation and AI sits at CEO level minus two. It's not a back office function. It's really important to our strategy. I'll give you another stat that is really different in our case. 70% of the EBIT impact that we have had through automation this year is new revenue. This is not about finding FTEs that we can remove from the operation. It's about scaling our operations. This is incredibly important. We want to transform the user experience. The interaction with an airline tends to be stressful.
The last thing you wanna hear: "Your flight is delayed." "Your flight is canceled." What if we try to help that using agentic automation to say, "Your flight has been canceled, but here are three options where we can rebook you. Would you like to choose one of those?" That sounds super simple for an end user. What happens behind the scenes is incredibly complex.
Mm-hmm.
By using these technologies, we think we can transform and obviously get a competitive advantage with that.
Thank you both. So I think we have time for one or two questions. I can't see name tags, but if you... We got a mic, too.
... Hi, is this on? There we go. Scott Berg with Needham & Company. Thanks for joining us today. We really appreciate it. How do you think about using platforms like UiPath in conjunction with these AI technologies versus trying to take the AI technologies and custom build your own applications, you know, through some other means to accomplish the work tasks that you're actually doing? I think there's a pretty big debate in our world around how much companies will use platforms like, you know, UiPath specifically to aid in these processes versus you all trying to do this on your own through some other, you know, internally developed tools.
Yep. Should I make start?
Yeah, go ahead.
Yeah. So we think that it is perfectly positioned to bring AI and add value. Let me tell you why. We did deploy a couple of years ago an AI team to do AI-specific solutions, and one of the things they were working on, for example, was delay prediction. Can we use all the data that we have to predict whether this aircraft is gonna be there on time, yes or no? When we try to deploy it in production, we struggle because in order to put it in production, as good as the model was, you need to think about the two steps that happened before the model is used and the two things that happened afterwards.
So we had to do a lot of development to stitch it together as part of the business process. Where UiPath excels is that all these technologies, including LLMs, et cetera, it's already part of that orchestration, so it makes it really easy to say, you know, automate the process using AI, not to think about AI in isolation. So I think that's, in our case, that's where it's been a differentiator.
I guess in terms of vision, similar, the reality of the world in companies is that we are not running business processes on single tools. It's not a single ERP, it's not a single CRM system. It's a combination of many systems. It's a combination of many manual processes of decision. So yes, I think many companies or many other vendors or ERP companies will work on bringing an AI experience within, I would say, that tool itself, but we still need to orchestrate it. We still need to stitch it together. We still need, I would say, to integrate that into a wider enterprise business process.
And at the end, I think you still need to have something in the middle that can help you, that can bring together all of those technology, and I think you are very well positioned in that challenge, which is an interesting one.
A Swiss Army knife, perhaps.
A Swiss Army knife.
Yeah.
Yeah.
Time for one more question.
This is Jake Roberge with William Blair. I guess to that last point of the Swiss Army knife and playing Switzerland and orchestrating all of those different agents together, like, in practice, when does it make sense to use the agent in a Salesforce portal or the agent in an SAP portal versus a UiPath agent that's orchestrating on top of that? Can you just help us understand how that looks in practice?
I will give my point of view, which is I don't think there's a unique answer to that question, first of all. It will be down to each use case that you need. I think ultimately, an Agentforce from Salesforce, you know, will be very powerful and bringing intelligence around how to optimize your lead, which is the process within that. However, the lead is only a part of that sale process somehow. So then how do you orchestrate that lead has been operating this way, then you want to take a further action into a planning system, I don't know, or you want to take another action into an early warning somewhere else? So that it's orchestration, I would say, across the enterprise, I would say, across departments, which I think will set the kind of boundaries between the various technologies.
Yeah.
That's my personal point of view. Again,
No, no, in my case, it's very similar. The way we think about it is if a product like Workday or Salesforce, ServiceNow, have this agentic element for it, we may consider it only if the business process lives within that application.
Yeah.
If it needs to talk to different things, especially the systems that we have, some of them developed in 1932, then it becomes really difficult, and we don't want everything connected to everything. So when it becomes a multi-system process, then we tend to use UiPath for that.
Thank you both for spending the time with us. We very much appreciate you guys making the trip. I think now we hand it over to management Q&A. Thank you both.
Thank you.
Thank you.
Thank you.
Thank you so much.
That's all right. Thank you.
Thank you.
Okay, I think this is time for open Q&A. Do you have a mic? Jake, do we have another mic?
Yeah.
Okay. Jake, maybe if they have two, we can do here.
Perfect. Thank you. Thanks very much for hosting. Looking forward to Forward this week. It's Michael Turrin with Wells Fargo. Good to see everyone. Daniel, you mentioned you've been on the road heading into this event for, I think, you said six weeks. Maybe you can just help us with where your customers are in that conversation around AI plus automation, if there are certain industries, certain use cases you're finding can present as the tip of the spear to kinda help get that vision out more broadly. Because we're dealing with just a lot of agent fatigue on our side. There's just, it seems like, a different press release every day. You have an existing base that's tied back to automation, so I'm wondering if you view that as an advantage and just-...
Things you'd point back to as where you see the ultimate opportunity coming into and out of this event? Thanks.
I think I talked to many different types of customers from like Tier 1 , you know, international banks, like Japanese mega banks, to like local telcos in, you know, countries in Asia to, you know, public sector in Dubai, for instance. I think the maybe there is a agentic fatigue in our world, but it's not in our customer's world. I think in all fairness, nobody's succeeded so far deploying real, tangible LLM use case in as part of enterprise workflows, and the interest is major. There's been a lot of money invested in the technology. They wanna see results. I think our message really resonates very well. They understand the idea you have to it's good to have a Switzerland-type approach to different business platforms. You don't wanna put all your eggs into one basket.
They understand the key to ground the agents with robots. Our robots work to bring information, relevant information, to the agents, to provide actions to the agents, but also to provide some guardrails, to even validate the outcome of agents. It's a powerful message, and many. We also understand it's the beginning of a phase. It's a bit similar with the beginning of RPA, when you have to find what are the best use cases, experiment, co-develop with customers, but the interest is so real, it opens all the doors.
Sanjit Singh with Morgan Stanley. I wanted to go back on the concept of what gives UiPath the license to win, and maybe it's around. I think we've heard the word orchestration multiple times. Is it in the tooling? And how do we think about the policy layer? I'm a little bit skeptical in an organization like Enterprise, that we're gonna have a lot of agents doing non-deterministic actions, right? Without that being seriously grounded or at least some human in the loop. And so where do you see kind of the sources of value in the stack, granted, with everyone having some sort of LLM capabilities or multiple LLMs?
So within the stack, where do you see the source of value that's gonna differentiate UiPath versus the, you know, multiple other competitors are vying for the agent opportunity?
Yeah. So let me start with our history, right? RPA is a technology that imitates people, and in order to run it reliably, we have to build this orchestration humans-in-the-loop part. LLMs are also a technology that imitates people. In order to run it, you will have this sort of expertise that we have, this orchestration platform. Second, you will have to grant LLM access to different third-party platforms. How are you gonna do it? Are you gonna give them the credentials to a third-party access? There is no way you will do this. You can expose yourself to adversarial attacks. You cannot grant a license to all the information in a system. So you'll need a sort of tool that is pointed and precise and give an LLM, an LLM only the information they need.
And when you provide an action, only an action that is reliable, precise, that can do like booking a flight. And, and then you, you'll have to control the access to these agents, like you can control the access to the robots. We have the platform that we are... And then the orchestration between agents, robots, and humans have to be there. And one thing that I want to point to you is, it's not enough to create one agent or one single automation. It's important to have hundreds of automations, and each of them would run thousands and thousands of times reliably. And this is, we have the experience. We have the best platform in the world to run reliably these automations. We are extending it to bring agents to deal with the unstructured part of the process. Our customers can take advantage of the existing automations.
This is, they can increase the automation program, knowing that later on, they can bring agents to even go after the unstructured parts of the process. Maybe Graham?
Sure. Maybe with an example, I can also help you understand a little. I said it before, the agents are only as good as the tools that you give them. If you give them a very powerful tool, they can do a lot of very scary things. Because the automations are built with governance in mind, and they're much more narrow, if you were to give an agent access to the entire SAP API set, for instance, it could do some very scary things based on what it finds in that API.
... as opposed to someone who's built an automation for a particular, a very specific function. It's much more reliable and much more controlled if you're able to more narrowly define that scope of action that it's able to do. And obviously, people have spent years developing those automations on our platform in a way that only a certain subset of people have access to, only have a certain access to a certain set of data, or in a particular context.
That all really matters so that it doesn't go off and I was talking to a customer about this agentic stuff, and they're like, "Oof, boy, we could really use agentic orchestration specifically because we are unable to do that." They told me a very scary situation where they had their financial data that was put into a bunch of Excel spreadsheets that were sitting on some SharePoint server, and the agent went in and changed a bunch of the data in there. They couldn't get it back. They couldn't. There was no change history. Like, they were stuck with some of those changes, and it was a very, very damaging.
I think that those kinds of instances really reinforce the need both to have the right tools through automations and to have an underlying trust layer, such that you can be sure that the agents are only gonna take certain types of actions and do so reliably and accurately.
Graham, I just do wanna emphasize one other thing. Sanjit, I think your question started with: What gives us the right to win? I think what's very important is you heard it from our customers as well. Like, when people are talking about Salesforce and Agentforce, that's within a process, right? And it's gonna be use case dependent. The amount of market that's there, we have a right to win a chunk of that market because we are going across all applications and going across the stack. Then it's supplemented by all the technological differentiations that Daniel and Graham talked about. So the right to win, we have a good right to win a big chunk of that pie.
Mark Murphy with J.P. Morgan. So Daniel, it's very exciting to, you know, hear you say that we've always handled structured processes, and now we're gonna, we're gonna move, and we're gonna handle the unstructured for powering, you know, autonomous, this kind of generation of autonomous agents. What I'm trying to understand is, I think the question we're gonna get is: Well, isn't that what the LLMs do? Aren't they providing, you know, the logic and the reasoning to pore over unstructured data, whether it's text, video, images, or anything else, and kinda do something with that in their own, you know, unique and special way? So I'm trying to understand, because the vision is exciting, but I just feel like that's what investors are gonna say.
Are you trying to move from being the arms and the legs up to kinda being the brain? Is that what you're saying? And then, if you're doing that, or do you just mean we can tap the API? You know, because I think people will say: Well, isn't that what OpenAI and Gemini and Anthropic, isn't that what they're doing? They're handling the unstructured data. Are you saying you're gonna tap an API and then kinda govern the process the other way so that they can't, you know, kind of contaminate data in the way that Graham described? But I'm just trying to understand what, y-y...
I'm trying to understand a little more clearly what it is that you've built. What is the technology you've built that will let you handle the unstructured data in a different way?
Yeah. Mark, let me, let me come with a, another metaphor, right? So there is brain, but, and there is the arm and legs and eyes, right? That are like the robotic equivalent. But there is also something very important, which is the nervous system, the nerves that connects the brain to the arms and to the command. This is something that neither us, neither LLMs have. So LLMs have the magic, the brain, to, when I give them the proper information, to, to call in to move the arms, move the legs, but this connection is not there. Nobody has this connection. This is what we are bringing. When I create an agent, this is actually what I am building. I am, I am wiring the agent, the brain, with the tools to do a specific use case, and this is really very powerful.
You cannot do it with OpenAI. You cannot do it with us independently. It's the combination. And when you create this new, it's a building block, and when you create this building block that is an agent, you have to manage it, you have to give access, you have to control the access to it. You need to have humans. You need to have agent interacting with humans. You need to make sure you deliver as part of this platform in a reliable way, and we have a lot of experience delivering automations in a reliable way. Does it make sense?
So you have built something in here that is the nervous system, and that's your-
Yes.
Which is, what was that?
... I think.
Yeah.
If you could tie it back to the demo ground?
Yeah, so I think to your point, we agree that the LLM is the thing that's being commoditized, and the models are being commoditized. So it's not there that their unique differentiation is. It's the layers below that support it. So there are a couple of different places. So the builder itself, it allows you to use any model you want to. And so our approach to the models is to be open and flexible. So it's very, very likely that in any given workflow, you're gonna use multiple models. 'Cause sometimes, you know, and even in our own testing, like sometimes Anthropic does better than OpenAI. Sometimes a specialized model that's specifically built for invoices, and your invoices, is gonna beat the pants off of any foundation model.
So the ability to basically plug and play the right model, and then, to Daniel's point, be able to connect those to the systems that make sense at the right time is a really critical part. So in my demo, you know, in the Autopilot demo, it used about five different models to accomplish that one scenario. It used a large language model for the natural language understanding. It used a specialized model to understand the information in that ID card. It used another specialized model for understanding what was on the screen to enter the data from that back into the screen, and it used an action model that helped us figure out what the next best action was going to be.
For each one of those, we have used many different models, even in developing that, and sometimes one model beats another. It is very likely that every customer is going to have an ensemble of models, if you will, that they will make use of for the right purpose, and those are changing all the time. The thing that we are differentiated based on is, as Daniel said, being the orchestration layer. The thing that brings the best model to the right thing, to monitor those models, audit those models, audit the actions that are being taken, learn from the patterns of that data, and then being able to construct end-to-end workflows that use the best model for the right the right job to be done, if that makes sense, and across all of these different systems and tasks.
Sorry, everyone, we're about up on time here. We will be having a cocktail session right after this with additional management, where we can have more product-focused questions.