Good morning and welcome, everyone. Welcome to EXL's 2026 Investor Day. For those that don't know me, my name is Andrew Toat. I am the new-ish head of investor relations and capital markets here at EXL. I see a lot of old friends and old faces here, and I see some new that I'm really looking forward to getting to know better. I was gonna start with talking for a second about the moment that we're in. When a new technology is introduced, there's a lot of excitement, there's a lot of opportunity that's involved with that, also with that comes a lot of noise. Our hope this morning is to cut through some of that noise and leave you with a clear picture of 3 things
It's how we view the opportunities in the market, EXL's strong positioning to help our clients, and how we plan to sustain durable long-term growth. Now more than ever, we're really excited to have you here this morning. A few logistics. We are live webcasting today, so the slides will be posted to our website today. We're going to hold questions until the end. After all the formal presentations, we'll have the whole group up here on stage, and we'll do a group Q&A. Coffee, as you know, is in the back, and we have bathrooms behind. Feel free to step out at any point as the presentation has no scheduled breaks, so we're going to go all the way through until lunch.
After Q&A, I would very much encourage you Lunch is going to be served through here, and we have the demos in the rooms over there of our IP and solutions. I would very much encourage you to take a look at those. Those are very important investments for EXL, which we'd love to have you have a live look at. Lunch will be served in the next room, and you'll have a great opportunity to spend some time with the presenters, and we have a lot of EXL leadership here. Safe harbor. As you'd expect, today's discussion will include some forward-looking statements regarding our strategy, outlook, and financial performance. Actual results may differ materially, and we refer you to the risk factors in our SEC filings for more information. Today's agenda.
I'll hand it over in a moment, but here's how the mornings will flow. Rohit will come up and kick us off with what EXL is seeing in the market, our vision and strategy. He'll be followed by Vikas Bhalla, who will discuss EXL's competitive advantage in data and AI. Andy will come up and talk about where we're investing in our data and AI-led architecture to create value for our shareholders and our clients. Vivek Jetley will then come up, and will bring it all to life. He'll give real-world examples of making AI real for our clients and expanding those relationships and driving business outcomes. Finally, Maurizio, our CFO, will come up, and we'll close, and talk about our strong and evolving financial model, and the continuation of our long-term growth.
Without further ado, I'll turn it over to Rohit Kapoor, our CEO. Rohit?
Thanks, Andrew, good morning, everyone. Morning. Thank you so much for joining us today this morning. We really appreciate it. My God, this room is full. It's an exciting time for all of us, and we'd love to be able to share with you what we are seeing in the marketplace, how we are thinking about playing the game, and what it means for us, our clients, our shareholders, and our employees. First, our ambition is to be the strategic trusted partner for our enterprise clients, helping them adopt and implement AI in their businesses. That's our simple goal. Our promise to our shareholders is that we'd like to be able to deliver sustained market-leading growth of revenue and profit.
We are in a fortunate position that we've been able to accomplish both of these objectives, and a large part of this might be a bit of luck, but a little part of this is some of the strategy and some of the thinking and some of the steps that we've taken to position ourselves to be in that kind of a place. I'd love to be able to share how we are thinking about our business and driving it. There are 3 key messages that I'd like to focus on today. Number 1, what is the AI opportunity that we see with our clients, and how do we see this evolving? 2, what is it that it takes to be able to be successful in this environment and with all the change that's taking place?
Finally, I'd like to close by, how is EXL specifically positioned in the marketplace to address this opportunity with this capability set and take this forward? Now, the one thing which AI has done is it's created a lot of excitement, but it's also created a lot of uncertainty because every single day or every single minute, there is a new, you know, headline news that comes up. It basically is something which all of us read simultaneously. The natural human instinct is to think about that headline and extrapolate that headline indefinitely and make the assumption that everything will work seamlessly from there on. The reality is that AI is a very, very powerful change that is coming about. The functionality is very real. The new capabilities that are being added on are really impactful.
At the same time, in order to be able to get to the real business outcome, you need execution, and that is complex, it is difficult, and that is something which only a trusted partner can help solve for an enterprise client. The other thing we hear about a lot is, "Let's throw a lot of compute at the problem." If you throw a lot of compute, then by brute force, one will be able to solve no matter what the problem is. The other aspect is, let's take the foundational models and there will be foundational models that will be able to do everything for every industry, every client, every use case, and it can be applied across the board. Therefore, the power of the model and the power of compute will solve the problem.
Our viewpoint is you need that, you need that compute capacity, you need the power of the model. Unless and until you apply the knowledge and mastery on data and you bring together the data and make it ready for AI, unless and until you take all of the knowledge about the business and apply the business context in the modeling of the data and the AI, and unless and until you use your expertise to fine-tune the AI and you deliver trusted execution, you cannot get to the outcome. There is talk about everything is gonna get disrupted by new native AI entrants, and that's happening. The startup ecosystem is absolutely galvanized.
You know, there are a number of startups that are now approaching the problem of how can you leverage AI in the workplace and be able to make a difference in terms of business operations. I think what the new entrants bring is a lot of creativity, they bring a lot of energy, and they act as a catalyst. I think, that's very, very real, and the change that they're making is very, very profound. Unless and until you can take that creativity and actually mitigate the risk associated, particularly with a regulated enterprise, and be able to do both, which is to take the creativity of the new technology, take a existing business model and disrupt it, and be able to manage the risk and mitigate that risk, you cannot really get to the outcome.
Our viewpoint is for the enterprise, risk mitigation is not an option. It is not that I'll try things out and if things break, so be it. It is about how do you move down that pathway and that journey in a manner that mitigates the risk simultaneously while delivering the profit and the benefit. Finally, this is my favorite, which is autonomous AI will result in complete elimination of roles and jobs and therefore there will be no work left to be done by humans. We all know that there are two sides to this. There is an elimination part of it, there's an augmentation part of it. Our viewpoint is that yes, a number of jobs will get eliminated and that will result in compression.
AI fundamentally is going to be augmenting humans and therefore what we will see is actually a much bigger TAM, a much bigger, you know, marketplace for us to be able to play in. The relevance of human intervention is going to get elevated towards more complex tasks, higher value-added work, and more judgmental capability. The other thing to keep in mind is the body of work, you know, in the first case with the in the hypothesis that autonomous AI will eliminate all roles, makes a very simple assumption, and that is that the body of work is a fixed quantity and a fixed amount of work. That is absolutely incorrect because as pricing drops and as the intervention becomes a lot more capable, the body of work actually expands significantly and moves towards higher value and higher complexity.
That's where we would hope to play in. I share these four signals with you because at EXL, again, if we have to drive sustained long-term growth, for us it's really, really critical to understand what is a fad, what is noise, how do we distill that noise and come up with a signal and a direction in which we want to take the company, and what is a secular trend that is taking place? Our goal is to position ourselves to be in the right place for that secular trend and to be able to read the signal correctly. If we can do that, we will be successful for our clients and our shareholders simultaneously. At the end of the day, our conclusion really is AI is a extremely powerful secular change that's taking place. It's gonna play out over the next several years.
This is the most important secular trend that's taking place at this point of time. At the same time, the only thing the client is concerned about is being delivered value and being delivered a business outcome, and the only way to deliver that is through trusted execution. That's what EXL stands for today, and that's what we hope to be able to leverage. Let me talk a little bit now on how we intend to deliver the value to our clients and provide you with a little bit of a deeper understanding of our approach towards delivering value to our clients. First, in our viewpoint, there is no application of AI without transformation that can result in extreme value being delivered to the client.
If you simply try and do a plug-and-play of AI, actually the value is very, very limited, even if you get it right. You have to apply AI and transform at the same time to be able to deliver outsized value. Second, in the past, if you think about outsourcing, that used to be done on the basis of a task. Tasks and parts of a process would get outsourced. If you think about the application of technology was applied towards elimination of friction points. In both the situations, it was a part of the process that was being impacted upon, whether by humans or by technology. In the world of AI, well, the first thing you've got to start with is you've got to have your data foundation correct. That's something which most, if not all, organizations do not have in place.
We think that that's gonna be a huge activity that needs to be undertaken. Just getting the data organized so that AI can be applied is a critical first step that every organization has to undertake. For us, what that really means is helping our clients manage both their structured data, their unstructured data, bringing it together from disparate systems, being able to work on any data that's existing in the enterprise and being ready for AI. The second part of it is the algorithm and the AI and the model. Again, the first time that you apply AI, the accuracy level is actually very low. You know, in our experience set, the accuracy level of a first-time application of the latest AI model is 60%-65% accuracy. You just cannot operate at that level.
We need to be able to elevate that accuracy level and take it into the mid-90s at a minimum for it to be effective. That requires fine-tuning, that requires iteration. By the way, that AI model needs to be continuously monitored. It needs to be continuously supervised, because otherwise it results in a drift taking place, otherwise you end up going outside of the guardrails and your governance, your security, your privacy, the risk associated with the accuracy of that model continues to fade away. That's a skill set and a mastery of how do you really leverage AI that's really, really critical. We are fortunate that we invested in analytics way back in 2006, and it's given us this tremendous capability of being able to be masters and be experts in AI. Finally, none of this works.
You know, you can be an expert on data, you can be an expert on AI. If you do not have the contextual knowledge to apply to both to data and to AI, this doesn't work. What that means is years and years of investing in the knowledge about our clients' business, knowing about their operating processes and their businesses, and that becomes critical. We believe that with AI, for the first time, the change has to be driven from the business side. It is not a technology-driven change alone. It is a business-driven change that uses technology, that uses data and AI and brings in the context, and that's what creates the magic.
Finally, if you are gonna transform the entire journey, you need to have ownership of the full stack. So you need to have capability of being able to impact data, AI, have the knowledge of the context, and be able to transform that entire journey. What's happening with this is clients are now engaging with EXL not for a task, not for a friction point, not for a piece of process, but they're saying, Take the entire end-to-end journey and transform it, change it, and provide me with the business outcome. That results in a significant expansion of TAM, a significant expansion of value, and that's where we are playing. We've been presenting to you for the last three years.
Every single time we've come to you, the TAM has increased, and the pace at which the TAM is increasing is actually really much, much faster than what we were seeing previously. There are 3 fundamental reasons why we believe that the TAM is increasing. Number 1, our clients are spending a lot more on data and AI. I think there's a report by McKinsey which talks about the spend on data and AI as a proportion of the spend on IT has gone up from 4% to 16% in 2 years. It's a massive change that's taking place, and it'll continue to increase. That spend is increasing rapidly, and that's the space in which we play in. When we talk about AI services, AI solutions on data, on AI platforms, that's the space that we play in.
Second, there are a number of adjacent capabilities that are kind of coming in together, and just like I spoke to you previously, managing the full stack means that we are actually addressing a much larger business opportunity. That's expanding it, and for us, in the past, we would deal with the Chief Operating Officer, and we would deal with the business processes. Today, we deal with the Chief Operating Officer, but we also deal with the Chief Information Officer, the Chief Data Officer, the Chief AI Officer, and by the way, the Chief Risk Officer, the Chief Marketing Officer, and the Chief Executive Officer. Our buying centers and our relationships have expanded very meaningfully. The third piece is it's no longer just the large clients that are adopting AI.
Today, if you're a mid-sized client, this is the best opportunity for you to be able to leverage AI and play in a democratized playing field where you can compete against the larger players. The mid-size players in the past didn't have adequate size and scale to outsource or to work with partners. Today, actually, that's become even more relevant. If you are a startup, you again need to work with partners because as you scale up your business, you need all of that support and all of that change management that needs to be, you know, undertaken alongside with you. What we are seeing is that the customer set is expanding very, very significantly and meaningfully for EXL, and we're gonna talk to you about a few examples a little bit later.
Why is it that EXL is one of the few players that's been able to demonstrate this sustained market-leading growth for several years now? Our viewpoint is we've delivered and proven the performance, and the one goal that we've always kept for ourselves is we're going to grow our profit slightly faster than our revenue, and we're going to grow our revenues at a market-leading pace. We've delivered that for the last five years, and we've also delivered that in the first quarter of 2026. Our portfolio today is in the Goldilocks zone. Having this mix of 60% of our portfolio being in exclusively on data analytics and AI allows us to be able to position ourselves for high growth, high value pools, and that business for us is growing very rapidly.
At the same time, the digital operations part of our business is 40%, and that allows us to learn, allows us to develop context, and allows us to be able to invest on the data analytics and AI side. Vikas will talk a little bit more about our portfolio as to how this is resonating in terms of actual operations with our business. I don't know if you realize this or not, because for us, we've been investing in our own intellectual property and proprietary assets of our own. Today, 25% of our revenue is based on EXL proprietary assets. Just think about it. When clients engage with us today, a fourth of that business is being done on EXL proprietary systems, EXL proprietary technology and assets.
This creates stickiness, it creates higher value, it creates an ability to expand margins, and it allows us to be able to build and grow our business at the same time as, you know, our ability to make change and make transformation. Finally, we have a delightful set of clients. Our clients have a extremely high NPS score, and that's very, very fundamental and foundational. Over the last 27 years, one of the things that we focused on is we wanted to be a highly customer-centric organization, and our viewpoint was, if we can make our clients succeed in the marketplace, we will automatically succeed. That's that extremely loyal, strong franchise that we have in place. It's in highly regulated industries, which is complex.
We like the portfolio of clients that we've got, and there's a tremendous room for expansion for us. We made a number of investments, and we are kind of increasing our investments. We've, you know, taken up our investments by almost 4 times. We've got a number of patents. We've got a number of, you know, technologies that we own, so we will continue to innovate in that. Some of the acquisitions that we have done in the past have been highly strategic. We still keep talking about the acquisition that we made in 2006 because in 2006 nobody was thinking about data analytics, and we acquired Inductis, and it's become the foundational capability of our data analytics and AI business today. We've invested deliberately in data management.
We did three acquisitions in data management, and we think data management is gonna be huge. You know, we are literally just scratching the surface of the work that we are doing on data management. Our Payment Integrity business was another business that we invested in which kind of brought everything together, and that's been growing really rapidly. Some of these acquisitions that we have done are highly strategic, and they are really creating the right kind of capability set for us to drive our business going forward. I will say this, that we will continue to do strategic acquisitions as we go forward and continue to build up capability. Finally, AI actually becomes even more relevant, and talent is a critical ingredient for ensuring that.
We have always invested in talent. We continue to invest in our talent. That's very, very important for us. We're making a slight change to the way we think about talent at the front end. We now think about this as being human on the loop as opposed to human in the loop. For us the difference is the human is providing judgment, the human is eliminating risk, the human is ensuring that everything is going as it should go, and there is adult human supervision in every single journey that we undertake. That's become really important.
The second hypothesis that we have is enterprises are going to move towards creating large language models for themselves, and therefore there is gonna be this need for having RLHF where we will train these models based on human learning and be able to do red teaming exercises and improve the models, do model evaluation, be able to put in context into the models, train the models, and therefore there's gonna be a whole body of work that needs to be undertaken that applies that human knowledge and applies it to an enterprise model. That reinforcement learning through human feedback is gonna become extremely important. Third, you all saw OpenAI's announcement day before yesterday, right? OpenAI deployment. You're gonna have forward deployed engineers.
Even OpenAI is doing that, Anthropic is doing that, everybody is doing that, which means even if you've got great technology, you need somebody to go in and help implement that. That's something which we are investing on our own and building up that capability. Finally, I just want to clarify, AI engineers versus software coders or software development talent is very different. Software development and coding can be done by AI in a very, very efficient manner. AI engineering and orchestration is not possible to be done by AI, at least not today. We are investing very significantly on that engineering talent to be able to stitch all of these capabilities together and deliver the outcome to our clients. Finally, in terms of our ecosystem, we've got a great partnership ecosystem that's been established.
Two years ago, this was very limited, but today these are very deep and strategic partnerships that we've created. I'll just highlight one of them, which is with NVIDIA, the world's most valuable company. The world's most valuable company this year in March steps up and calls EXL the Advanced Technology Partner of the Year. We are really proud of that recognition. NVIDIA, by the way, sees everybody, sees all the players out there, yet they choose EXL as the Advanced Technology Partner of the Year. That's only because they see how we understand the business context, how we understand and apply data and AI into the business, and we deliver real outcomes to our clients. Nothing happens by chance. I think, you have to be very deliberate about how you build an organization.
For us, creating EXL over the last 27 years has been a very fulfilling journey because we've always believed that if we have the right leadership talent, what that will do for us is it will create the right culture in the organization, and it will allow us to handle the most change, most volatility, most uncertainty in a very certain way. I am really, really proud of this team because this team is experienced, this team collaborates, this team works together, and what we've got is such a high talent density that we can undertake any obstacle that comes our way, which is not known to us even today. I think, we are really well positioned for the future with this team.
I would like to call out. We've just made one addition to the team, just, you know, this week, and that's Bhupender Singh, who's joined us as the President and Head of International Growth Markets. It should just show you, number one, that we continue to add to our talent density and bring on strong leaders in the space and be able to kind of execute and drive our business forward. It also should, you know, indicate to you that our emphasis on the International Growth Markets is a key priority for us. We've taken that business up quite significantly over the last few years, our ambition is to continue to drive that much faster. With that, our blueprint, I'd just like to summarize, is very simple.
Number one, we believe data, context, and AI is what allows us to create enormous value for our clients. What is needed is trusted execution. EXL has been able to demonstrate that repeatedly with its customer base. What that trusted execution results in is a very high level of business outcome on a sustained basis for our clients. If we deliver that outcome to our clients, I think we will continue to have durable growth and profitability. As long as we focus in on these basics, we'll be all good. Thank you very much for my session. I'm gonna pass it on to Vikas. He's gonna take you into more detail on how we apply data, context, and AI in our business and how we apply execution in our business. Thank you.
Thank you, Rohit. Good morning, everyone. It's wonderful to be here today. The 4 key messages that I will be focusing on over the next 20-odd minutes. Number 1, a little bit more detail of what Rohit spoke about. That true value for AI in the enterprise comes from data, context, and AI with trusted execution, and how EXL brings those components very nicely together for the enterprise. The second is that to be able to accelerate these for the enterprise, we have created intellectual property and agentic platforms so that we can do this thing at speed and scale.
We'll also talk about our 2 businesses, which is data and AI and operations, and how they are now working in a very symbiotic way from a viewpoint of data context and AI, and how that will allow both the businesses to grow nicely together. Finally, we're gonna talk about the 4 key demand vectors that we're actually seeing from our clients. Before we get to a little bit more detail on data context and AI, let's take a step back and look at the evolution of AI over the last, you know, 10-odd quarters. Q4 2022, that is the time that generative AI was unveiled. We have seen progression of that over the last, you know, 2 and a half years.
Initially it was the first time that we could actually have AI, which apart from just analyzing, could synthesize. Could actually talk to us, create summaries, create content, create output which was human-like, which is very exciting. Then progressively, we learned as to how to make that more intelligent, sometimes by infusing that intelligent in the model, but many times by creating systems and structures around the model, for example, RAG models, so that you could become more intelligent. We also started moving into coding, and we found that the ability for AI to write and to edit code actually is pretty good. I'm sure that you know that's one of the most talked about and relevant use cases.
Finally, late last year and early this year, we started thinking about taking AI from just decision-making to action, and there was this thing about bringing that to the desktop. However, when you look at the enterprises over the last same time period, their journey has been a little bit different. When the excitement started with AI that could actually articulate and create content, there was a lot of excitement, and everyone started jumping into use cases. We found multiple use cases coming up. Everyone was experimenting. A lot of lab work was happening. We did not see almost any production-grade, scaled-up use case. Lots of experimentation, but no real production. The enterprises realized that one of the things that can be done is to take these and give it to the colleagues, to the employees, and tell them, you know, improve personal productivity.
Desktop applications started coming in, summarize emails, help draft emails, organize calendar. That phase went through. That is because every time we were taking AI to try and change a core workflow, every discussion on AI was very soon becoming a discussion on data because everyone realized that data is not in a place, it is not fit enough to be used for AI. We saw this period where most enterprises started working extensively in modernizing their data stacks and making sure that data has more meaning, which is very important for AI to be effective.
It is only very recently that we have actually started seeing that organizations now are beginning to ask the question that we need to move from experimentation to production, which is in select but in core business operations, how do we infuse AI, but infuse AI at production grade and at scale. Clearly that shift is happening. There are challenges, significant challenges. This is not a drop-in technology. You can't actually take an AI model and say, let's just drop this into an operation, and it's gonna become AI. What are the enterprises looking for? What are our clients looking for? The first thing they look for is it creating customer and business impact? Now, the customer impact metrics have not changed. They're still the same: customer satisfaction, NPS, time to market, responsiveness, and so on and so forth. The P&L metrics are very clear.
Using these, can I actually make better market impact? Can I grow faster? Can I make more money? These questions still remain because no point just having a fancy tool unless it can create a real customer impact and a business impact. The second thing that's happening is that because we have seen a lot of experimentation happening, it is time that we start seeing some scale-up. We start seeing scale-up in not everywhere, but select core business operations. If a healthcare payer, a core business operation is claims. It is member services. Can AI create that impact there but create that impact at scale? Finally, trust. One of the things we have seen is that the decision-making is moving from deterministic to probabilistic, which means that there's not a formula which is driving decision-making anymore.
Ability to create audit trails and make sure that you can give evidence of why a decision was taken the way it was taken becomes very important. To be able to do that, what we're finding is that 3 important things need to happen. First, data needs to have meaning and needs to have access. Second, context needs to come in so that it is relevant. Third, we need to scale up with speed. 3 critical things that need to happen. Let me just talk about how these things are brought by EXL and how we are able to create this value for AI in the enterprise. The first element is data. There are 3 important things with respect to data.
The first is you need speed to access. Second, you need to have the confidence that it has the ability to manage multiple kinds of data. Data today is not only structured data. It is structured data, unstructured data, internal, external, industry, multimodal data, and multimodal data sometimes with high velocity, video feeds coming in. When you manage all of these things, the old infrastructure, the old archaic infrastructure is no longer good enough. Organizations are working towards modernizing their data stacks. Rohit spoke about the data management capabilities that we have built up over the last many years, including the work we do in analytics and the data management assets that required. We are doing extensive work with our clients to help modernize their data structures.
For example, if a large insurer today has ambition of actually converting their claims and underwriting to agentic, we have a massive engagement going on with them just to fix their data and take that to a modern stack. The second element about data is data has to have meaning so that AI can be effective. For data to have meaning, there are two connections that need to be made, and it's really important that those connections are made. First is that you need to have the lineage, which is understand how the data is flowing in the organization. Where is it starting from, and where is it ending, and what journey is it taking? That connection is important. The second connection which is important is that data element has to be related to other data elements.
For example, if there is a claim, the claim needs to be connected to a medical record, a customer, a customer profile, a contract, a policy document. That graph, which is called a knowledge graph, is extremely important. This lineage and knowledge graph is what gives meaning to the data. When AI is deployed, it can be effective. The last one is that data has to have trust, because if decisions are gonna be made based on purely data, then you've got to make sure that it's trustworthy. Governance and quality, which is not by the way a one-time activity, it's an ongoing activity, is important. EXL, through its rich understanding of these domains, is able to bring this context. We understand insurance, we understand healthcare, we understand banking.
We bring this context and we bring this trust, and we have created a platform that Andy is actually going to be demonstrating of how we're bringing all of these things together for a client. The second one is context. Sometimes this question is asked, "Oh, is context what was being called domain earlier?" Yes, but context has more elements to it. The first element context has is the industry domain. It's about the customer segments, it's about markets, it's about products, it's about dynamics, it's about regulation. Everything which is around an industry is context. What is also context is how an enterprise works, which is very, very specific to that enterprise. Their workflow, their systems, the way that they are organized, their policies, their procedures, their customer preferences, their strategies, all of that is also context.
If you think about EXL's ability to bring context both at an industry level, because of the sheer focus we've had over the last many years, as well as at an enterprise level, so our deep, long relationships with these clients allows us to have an understanding. Just to give you an example. For some of our clients, as they have been through technology changes over the last 10 years, we as an organization are the ones who've been through change management. Our understanding of their workflows and policies is far richer than even what they have internally. Our ability to bring industry context and client-specific context is very, very high. Finally, to be able to give speed to it, you've got to bring some accelerators. Now, these could be spot accelerators like extraction engines or customer assist agents.
They're also deeper and broader like, you know, fine-tuned LLMs, like agentic platforms. This combination of ability to modernize data, to bring meaning to data, to bring trust to data, to bring the context which is at an industry level as is a client-specific level, and then to bring accelerators is what is allowing EXL to be able to create this value for the enterprise in the field of AI. Nothing gets done if you do not have the right talent. Now, in the 65,000-odd colleagues we have at EXL, we have over 17,000 who are data scientists, data architects, AI engineers, solution engineers, business architects. These are people who understand, and by the way, they've been working on these domains, these industry domains for a very long time.
Their ability to bring all of these things together at speed to drive a trustworthy execution is extremely high. One element of actually bringing success to the table is talent. The other element is that are you all the time bringing a pure services model or are you building some IP and platforms to make it more successful and do it with speed? That is where we have invested in creating agentic platforms which allows us to work on AI in the enterprise at speed. There are 3 platforms and my colleague Andy is going to be walking us through a case example for each of these 3 platforms for you to get a little bit better understanding of how this actually works. The first one is all around data.
The mechanism of modernizing, creating meaning of the data, the lineage, the knowledge graphs, the trust. We have created an agentic platform so we can do this thing at speed, at scale. Now all the data management work we are doing for our clients is all happening on this platform. That gives us a significant advantage of creating this value for clients at a rapid pace. The second is around decisions. There are certain decisions that you cannot leave to AI. Well, at least not right now and not in the foreseeable future. These are critical business decisions. I mean, let me give you an example. If a critical claim is declined by the use of AI, we may not be ready for that today, we may not be ready for that tomorrow.
You still need to bring in models that you actually created using data and analytics. To be able to deploy those models, we've created an agentic platform. We have an agentic platform which now allows us to create decision models and embed them as part of the processes very, very quickly. Finally, EXLerate.ai to rapidly deploy agentic platforms. It is one thing to create a technology platform because you could argue that if you bring significant technology capability with the ability of actually working with agentic systems, you could do that. What you'll find here is that this also comes with 27 years of deep domain expertise in these industries, 20 years of data analytics and data management capabilities, and 3-4 years of core investments in AI. As a result of which these platforms come with pre-built agents on the domain.
They come with ontologies for the domain. They come bring in context graphs. For example, today in a claim cycle, we already have pre-built context graphs, which are available, which can be deployed very quickly. In a sense, the platform is a combination of our domain expertise, our data expertise, as well as our ability to create agentic platform. You'll find this demo very interesting that we will do in a few minutes. Let's just change the subject and talk a little bit about how the two businesses, which is operations and Data and AI, are working symbiotically. For that, I'm going to use an example, let's use claims because I think it's a very effective example to explain that. On the operations side, we work on claims processes for clients.
We do claims lodgment, we do claims processing, we do exception management, we do claims customer service. We also do post-claims works like reserving. We do subrogation work. We basically manage the complete claims operations. On the data analytics and AI side, we do customer segmentation work, customer value work, retention work, fraud models, catastrophe models, and we actually help clients in working on their data modernization, data semantics. All of that work actually happens there. We have discussed context is important, but remember, context cannot come only from one of these two elements. It has to come from both because they are a bit different. From the data side, the context which comes in is insights. The context which comes is customer segments.
The cost context which comes in is the response to certain variables and how that entire thing works from a claims perspective. All of those insights and intelligence actually comes from data and AI, which feeds our operations and makes that smarter. There's also a context which goes from operations to data and AI, and that context is around ontologies, semantics, understanding of customer personas, the way that the workflow is working, the way that the controls work, the guardrails, the regulation. All of that context actually flows from operations. If you think about context, these two businesses actually are working beautifully together. They're feeding each other. What we are finding is that this engine is working very nicely for us.
Our ability to be effective in data and AI is being helped by operations, and our ability in operations is being helped by data and AI. I gave you a claims example, but the same example is relevant because remember, we do data science, data analytics, AI work for clients, and for the same clients, we run their operations, and it's basically a very symbiotic relationship. As a result of which, we are finding that both our businesses are growing, and I'm sure you have tracked results for the last few quarters, including the most recent quarter that we actually announced. The data and AI business is growing. We spoke about the reasons. Expanding TAM. We bring a differentiated value proposition of data context and AI. We have the talent pool, which is large, diverse and expanding, and we have the platforms to bring speed and scale.
If you look at the operations business, our viewpoint is that there is significant headroom for growth based on this differentiated value proposition. First, the penetration levels are still low. I mean, different numbers, you know, float around, in my assessment, and happy to have this conversation offline, I think it's about 20%-25% right now, max. That's the level of penetration, which means that's the level of outsourcing which has been done. Significant headroom for growth. Number 2, most of the work we do is in complex, regulated and very, very sensitive operations, which are deep in the domain, and you need expertise to manage that. More importantly, if you have to transform them, you need this expert-expertise to be able to transform. EXL working on those operations makes sense for the clients.
Third, both large enterprises and medium-sized enterprises actually are finding that it makes sense to outsource. Large enterprises traditionally have a more mature outsourcing engagement, but there are certain parts they were never outsourcing because they thought that they were too close to them. There was some uniqueness or there was some complexity. Now that they know that outsourcing to a company like EXL will automatically help infuse AI, they're much more open to these functions being outsourced. Medium-sized enterprises who were not necessarily the most mature outsource earlier are realizing if they have to basically be competitive in the world of AI, they may not have the ability of actually infusing their business with AI. Working with a partner like EXL and having them run these operations actually would help them with this journey.
We are finding increased demand coming from both large enterprises as well as small and medium enterprises, and our viewpoint is that this will help in a symbiotic relationship that I just spoke about on the previous slide for us to continue to grow. My final slide in terms of the demand vectors. There are 4 kinds of demands that we're actually seeing from clients, and each of them are on a increasing path. The first one is data for AI, which is, like I said, every conversation on AI becomes a conversation on data very quickly. There are 2 kinds of demand we see.
One is a use case-driven demand. That is, if I'm actually working on my underwriting system and I want to make it agentic, I need to fix everything in the data around it, be it modernization, be it semantics, be it trust. EXL, can you please work with us and help us do that? I think Vivek is going to be talking about one use case as an example of that. The second demand we're seeing is at a foundational level. This is where enterprises are saying, "Okay, not for a specific use case, but we want to totally modernize this stack at an enterprise level. EXL, can we actually work with you, and can you help us do that?" Two kinds of demands, use case-driven demand and foundational demand. The second is AI services.
AI services is from helping clients understand this very complicated provider landscape in terms of different technologies and tools, but helping them from there to deploy agentic in their workflows. And deploy that, bringing in the components of data, context, and AI. There is demand which is continuing, saying, "Can you run my operations?" We just spoke about, you know, the operations business. "Can you run my operations, but can you run it AI first?" This has been on for the last 3 years, but it is changing a little bit. Earlier it was AI-enabled, which is run my operation, but then infuse some AI in it. That infusion is, you know, put an extraction engine here. Let's put a customer agent assist there.
It was, to some extent, a bit tactical and transactional, but an infusion of AI. From AI-enabled, it is now becoming AI-led, which means, "Can you please work on creating an agentic workflow?" Then there has to be human on the loop, and human on the loop means that somebody has to still manage exceptions, monitor model drift, do reinforcement learning. There always will be a human workforce which will be working with that, but it has to be agentic-led. Finally, these solutions, integrated solutions. One of the examples we talk about a lot is Payment Integrity business. This is where we bring all the components. We bring in the data, the AI, the technology platform.
Because we own the entire stack, most of these businesses are typically on an outcome-based pricing, which means there is value to be created for the client and there is value to be captured for EXL. Rohit gave us that number of about 25%, which is on EXL IP. This is broadly that segment, which is all of these solutions which bring everything. Another interesting way of looking at it is the first one is data, the second one is data plus AI, the third is data plus AI plus operations, and the fourth one is data plus AI plus operations plus proprietary technology. What you will find is that data and progressively AI is basically becoming present in everything that we are offering to our clients, and that is where the demand is coming.
Vivek, you know, is going to be talking about these 4 demand vectors with some specific client case studies in a bit, and I think you'll find it very interesting in terms of what we are seeing. With that, I'm going to give it to Andy, and Andy is gonna talk about how our data and AI architecture is helping this accelerate for our clients. That platform that I spoke about, how that platform of EXLerate.ai, EXLdata.ai, and EXLdecision.ai actually is working with a use case. Thank you very much.
All right. Thank you, Vikas. Good morning. By the way, there's a side bet going on that I'll not finish on time. Please give me verbal nods, be attentive. If you're not, then I'll be making more stress in explaining. That'll take more time. Vikas talked about data context and AI and the trusted execution part. Rohit also talked about our patents and differentiation. Both highlighted the platforms that we are building. What I'm going to do, firstly, maybe take an attempt as to why enterprise AI adoption is complex. Number 2, most importantly, how are we making that easy and helping customers deliver value? 3, most importantly, just bring it to life with a demo and bring all these concepts together.
If some of these terminologies you didn't follow, hopefully the demos will make it clear. That's the simple plan. Just one caveat, it's an architecture conversation. As much as I try to make it non-technical, I may not be very successful, so just bear with me, but that's the world we live in today. You heard data, context, trust, reinforcement learning, AI engineering. All of those also became priority for us for capability investments. What we did is think about the enterprise value, and one spectrum was, how do we make production-grade enterprise way of delivering this value? Two, where are we differentiated that gives us a leg and perhaps customers can see that value more visibly and it cuts across different clients and different areas we deploy? Rohit talked about patents.
Let me just double-click a little bit on how these patents just come about. Firstly, you may or may not know this, EXL has a large team of PhDs, researchers that actually come from Google, Stanford AI Lab, and other reputable organizations, plus with EXL engineers and domain experts. This R&D group is essentially always looking out for 3 things, and I'm just going to contextualize this with an example. First, when Claude Code came, a lot of people realized Claude Code is not just the Claude model. It's the Claude Code, it's the permissioning, it's the sandboxing, it's the tooling. We realized what happened to code can actually happen to enterprise workflow. That gave us the thinking to build agent harness for enterprise workflows, and that was 1 area of patent for us, and it's applying now everywhere where we deploy these.
Second thing, Vikas talked about knowledge graphs. Knowledge graphs, as Vikas articulated, is a living graph of these decisions, interconnected relationships, and meanings. What we realize is that organizations are using knowledge graph only for connecting the meaning and just looking at how different data sets are connected. What people don't have is every time I decide, where do I capture that? If you don't, downstream agentic and agent is not self-learning. That gave the room for our context graph patent. We are always looking at the latest and greatest research, we're always looking at what's happening, and then we are applying it to enterprise context. I just wanted to double-click on the patents a little bit. By the way, this is, actually a very dangerous slide, I'll tell you why.
Most CEOs and boards go to conferences, and they look at a five-layer cake or a six-layer cake. They come back in the enterprise, and they'll say, "Why can't I make and deploy AI real in my enterprise? All you need is infra, data, model, apply some guardrails and governance, build agents, and just there you go." That's what people go back with, and then teams are struggling, "Oh, my God, somebody wants something next week, and I can't deliver that." Let me just take a few slides to explain what it really takes to make this real. I'm just gonna, I don't know if it's a cake or a bakery. I'm just gonna try and explain it to you why it gets complex, but how are we helping the clients. Firstly, let's double-click on the infra layer.
EXL, as Rohit mentioned, has partnered with Google, NVIDIA, Microsoft, AWS because clients are not yet ready to give their keys to the kingdom to one provider. They need flexibility. Number 2, even more importantly, think about sovereign needs, think about healthcare, HIPAA, and high trust, and think about GDPR when you think international. It needs to be secure and flexible, so we partnered and made sure that this is dependable for clients to scale. I'm not gonna spend too much time on it. Vikas talked about EXLdata.ai. There is one element that I'll highlight. He covered 125 agents that we are using for faster data processing, making it ready for agentic workloads. By the way, per a Gartner study, this is still 50%, 60% effort. Let me give you another dimension, a newer problem.
World's most data, 90%, got created in last 2 years. Many companies have data, they don't know what to do with data. Believe it or not, that's a 70% or 80% problem, and that's what explains the context part of it, which we'll double-click in the demo. Vikas talked about knowledge graph, I talked a bit about context graph. Here is a third very important element for regulated industries which doesn't get talked about, and that's the symbolic, the Neuro-symbolic AI. You will use neural network for things that neural networks good at. In the enterprise, there are policies, there are rules, there are guidelines, and that deterministic logic, you have high trust. It does not hallucinate. You're gonna still depend on it. By design, this is Neuro-symbolic AI for us. This is not just EXLdata.ai.
Our entire architecture follows that principle 'cause it's very important for regulated industries. Let me just next click at very quickly at the AI model layer. EXL today accesses 44, 45 models by the count yesterday. Maybe 2 more got added this morning, which I don't know of. We access 45 models. Think about Groq, think about Nemotron family from NVIDIA, think about Anthropic, think about OpenAI, you name it. EXL has also created some specific domain LLMs, and couple of them have got patented. One might ask, "Why would you create domain LLM when models keep getting created left, right, and center?" Very simple reason. Sometimes model accuracy is not a factor of 89% versus 94%. Adjusters, underwriters will reject at 88%.
For higher accuracy, sometimes faster speed and better token economics, one other thing that I'm really obsessed with is token economics that may keep coming up because enterprise need to manage costs. We, at times, have to create smaller models because we believe they have a purpose and they have a need. Rohit articulated we see that need will keep coming up. By the way, it's one-tenth of the cost. I'm gonna click next on the agent harness. By the way, we could just spend an hour and a half on these 10 sub-boxes. If I were to just pick, let's say, one thing that just highlights how we've thought about these things. When OpenClaw was launched, you know, nobody will put this in the enterprise because of safety concerns and issues that OpenClaw brings. Understandable.
What people didn't see is OpenClaw is a marvel in engineering in the way it handles memory. It has 10 patterns of memory, and we learn from it. The R&D team went through the all the papers, everything, just took that code and just bifurcated it in any possible detail, and we applied those memory principles into our Agent Harness. Agent Harness for us is the agent orchestration in a persistent state where it doesn't lose memory. Why? Large, complex workflows don't run for minutes or hours, they run for days and weeks. In our memory principle, first, we make it accessible. You can audit it. It's not a black box. Many platforms have that problem. Secondly, you may or may not know this, AI has an amnesia problem. When the context gets full, it deletes, and we do context flush by design.
We manage our long-term and short-term memory really well. Thirdly, I don't know how to put it, but deduplication, which is we make it super lean for token economics. How? We don't want agents to learn what they've already learned once. Those are just three principles I'm talking. Each of these is like how engineering depth has gone into it to make sure agents can reason, they can run, they can execute, but they don't lose the state. EXLdecision.ai, this is something that you would not typically see in a reference architecture, but there's a reason why we do this. Vikas alluded to our analytics practice. 18 years of building machine learning models, pricing, underwriting, reserving, actuarial, clinical adjudications, and these 800+ algorithms that we've built over so many years, that expertise, we actually applied agentic to accelerate the model building.
If you want to use a machine learning model, which may be the right thing for you because of the reasons I mentioned earlier, hallucination control, you trust it's governed. Absolutely. We've brought all of that into EXLdecision.ai, and this, believe it or not, sells as much as probabilistic models today 'cause the value is still immense. Governance and guardrails, I just want to mention one thing. The newer world has newer nuances, and one of the big differentiation that hopefully will come out later in the demo is how we are applying very domain-specific guardrails. There are guardrails for inputs, how you manage outputs, which most technologies do. There are some very specific nuances which I'll bring up. Last but not the least, this is where the magic happens.
You bring everything together, you charge on outcomes, and you take a portion of value. Subrogation, Payment Integrity, bank transaction fraud, underwriting, Paymentor, which is collections, and Vivek will make some of this very real. Essentially, if you think about it, we've taken all experience, AI depth, data, and brought them in a manner where we can charge on outcomes to our customers, in most cases, as a percentage of value. As AI scales all the good work we do, the benefit still sort of accrues to us. Second thing which gets forgotten. Speed. In today's world, if there is one thing every CEO has on their mind is speed, and this brings speed, you know, something you can go 60% and rest of it perhaps tweaking is required.
We are looking at more and more inspiration of such ideas that we can create more agentic workflows. I'm doing good on time, Pam. All right. I'm just gonna tee up the demo and then I'll show you what the demo does very quickly. We pick claim for a reason as an example because, one, everybody can relate to it. Number two, you know, when you hear stories, claims is often talked about as, you know, you can automate the workflows very fast, right? The key point I just wanted to make here is that in this context, the demo that I'm gonna show you, let's look at three or four problems. Number one, 30-plus just systems. You look at third-party sources, unstructured data, medical records, police reports, all of that.
This is a very, very complex data environment. Number 2, fragmentation. Data is sitting in silos and in different places, they're not connected. The problem of that is sometimes you can miss an important context that can have an impact on your liability, recovery, or a fraud. You know, every claim has its own nuance. For auditors, for your compliance teams, they need a decision trail of every decision you're making and why. As you all know, claims have regulations. There are workflows that don't run for a day. Claims runs for weeks and months, right? It's important in any agentic design you don't lose the state. What did I escalate? When did I speak to Andy? What was the context? What was told to the customer? It needs to be in a state where it can remember that context and not lose its state.
Lastly, in all the claim design, you'll be looking at multi-agents, and they have to interact with each other and at various places, you'll be doing human handoff. I just want to articulate that you think of that simple layer of the cake and you think about what goes into it. This is sort of the reality. As Vikas mentioned, this is where I'm just going to bring the platforms together and I'm just going to do a quick demo after this. What you'll see is EXLdata.ai will help create, bring the data, establish the lineage, and build a context foundation. Decision.AI and between the AI model layer, there is a routing logic that, based on the context, accuracy need, and cost, will route the right context to the right model. That's one of the other differentiators.
We've created a routing logic because otherwise you have 45 models, how do you choose what to do when? Third, I talked about harness, which will come to life, governance and guardrails, and then EXLerate will bring this whole workflow together. With that, I'll just take a liberty to do the live demo, and I hope it works. I need my glasses too. This is Actually, let's just start with Data.ai. All sources you can see how the pipelines came together, what am I ingesting, what am I parsing, what transformations, and what validations, right? I'm not gonna go through all the things, I'll just highlight one important thing, which is our patent and differentiation. This is what Vikas was talking about with lineage. You know why this is different? By the way, this is compatible with Unity Catalog.
You can feed into Palantir Foundry, you can feed into Snowflake, not an issue. You can open source it to Collibra. Whatever you like, you can do it. The beauty of this is most companies will make this transparent to you in their platform. EXL will make the lineage transparent to you regardless of the platform. We'll bring the entire meaning flow state of the data. Two big advantages: One, the speed. Two, 40% of data estate in the enterprise is unusable, but they still go ahead and migrate everything because they just don't know what to leave behind. That's just one thing I want to quickly highlight on lineage. Let's pick one other thing. Actually, let's pick a use case so that you can see how this all works together. We talked about context.
Think about ontologies as the relationships in the business context. This is where the first time data marries context. Think about knowledge graph, a living graph, like Vikas articulated, of how everything is connected, a claims to a policy to a record to a call to a transaction, and what that really means because this is all the facts that you will need where agent will start to make informed reasoning and not just look up data. That's the big thing about it. This is one that I'm really proud of, the context graph, which is one of the latest ones. What we've done is, in addition to knowledge graphs, we capture every decision trace.
This was a research paper, now there are companies that have formed on it, but this is in our stack for the last 3 months and one of the most deployed things with the customer. I'm gonna make this live. Let's just look at what happened here. Let's take a case. I went to EXLdecision.ai. Based on the client recommendation, the model routed and said, "Pay the claim, but do not send it for recovery," because the model, the threshold is 97%, the police record is 99% certain, you don't need a recovery, but the case is okay to be paid, right? It fell in the queue of the agent. I'm just gonna go to the context graph and I'm gonna ask, "Why claim 1847 was recommended for subrogation?
Give me the evidence. Now look at the beauty of this. This is natural language, and this is for geeks, whichever you like. It got approved, and look at the magic of the context. Knowledge graph suggests there is a potential of recovery based on the case laws and comparative negligence. The cloudy weather and low visibility incident suggests there's a potential for a liability split. There is a Texas guidelines and arbitration ruling which otherwise it would have missed. Perhaps you can even pursue a split liability, or you could go for the entire thing. This just brought, by the way, how this all got applied? Just very quickly, this is my agent studio. I looked at my subrogation agent. I looked at the guardrail. This is the guardrail where I have all the details of the domain. Remember symbolic verification?
I have Texas Subrogation Statute of Limitations and all the other states. You may or may not know this, EXL handles 12% of subrogation of entire U.S. We collect $25 billion in recoveries for the companies. The knowledge of subrogation married with AI. Essentially, now I go to the trust and governance layer. I can see, yes, liability validated, subrogation done. Let me review the case. Human on the loop and not in the loop. Everything else is provided to you. I'm like Waymo. If car is now getting out of control, I'll just take the control of the car remotely. I'm just looking at what went into in, why did I do this, and okay, makes sense. Let's just approve the subrogation.
The beauty of ontology knowledge graph and context graph is ontology gives me the schema and the meaning and the domain blueprint. Knowledge graph then brings the relationship and living graph, and decision tree real-time applies the brain and the context for the decision. Everything is. You know, we could spend hours on this demo. We are looking at observability, token economics. We can do any kind of conversation and ask questions, and all of that is available in the system. That just gives you an example of what we've been able to build. One homework for all of you, please find me a technology, and maybe I'm a little proud here, but please find me a technology that can do all of this in the way I just demonstrated.
You'll probably find eight, nine different things will have to come together for this to happen. It can happen, but many things will have to come together. Why? What? Right? That's the important question here. You heard data context and AI. What we've been doing is we are creating IP. We are charging on outcomes. Obviously, every time we are delivering trusted outcomes and trusted better outcomes to our customers, this keeps improving. This flywheel, I learn every time. I trace every agent decision. The human pushes to the agent, "No, don't give this work to me next time." On the loop means what can I give to the agent? In the loop means every time give it to me, and I'll just keep looking at it, right? Here is a bigger story of this.
If I can charge on value, think about repeatability, think about every time I'm able to take it to the customer, and think about every time at the speed of deployment, how I can scale and multiply this over many customers on all those chosen workflows, right? I think, that's what I just want to leave you with. Thank you. Okay.
Thanks, Andy. I hope you guys got a sense of why did we win that award from NVIDIA for technology partner of the year, right? It has to be this complex, because if you don't make it this complex, you're not getting that award, right? We're very, very proud of what we showed you just now.
For those of you who are getting restless, we are into the home stretch now, so it's gonna go faster, and there are gonna be no more architecture charts from this point forward. It'll speed up. Look, my job's a relatively easy one. I just have to show you how it all comes together and how we're supposed to make AI real. I took a cop-out. Instead of me talking to you about what I believe, I'm gonna show you client examples. I'm gonna show you real-life examples. No pilots, no proof of concepts. Real-life examples of clients that are using our AI and our data capabilities in production, and that are driving phenomenal value through that. Real outcomes. You're gonna get a sense from this section about, number 1, why do our clients pick us?
Why does EXL win? What is it that we deliver for them? you know, what is it that is gonna create that sustainable value advantage for us? A couple of key takeaways. number 1, I just want to reiterate, deeply embedded within our clients. Rohit's talked about this. We work with regulated industries and have a phenomenal client base. It's probably our strongest asset. What you've seen from the presentations that you heard from Vikas and from Andy is we now have that full spectrum of capabilities across their entire data and AI needs, and we'll show you how that's translating into wins. I wanted to end with is basically the value equation for EXL, where we bring that together with the TAM, our capabilities, our IP, all comes together to create sustainable value for us.
Who do we work with? This is a page that we really like to brag about, because take a look at it. It's got 115 clients that are in the Fortune 2000. We've got 400+ clients that we're doing deep work with on data and AI, and this wasn't just a cursory check the box. We actually went back and checked what's the scope of the work that we do and what's it, you know, what's it delivering for the clients. The next part, our average client tenure is 10+ years. Take a look at the quality of what we are doing for them. You've heard all of these reports about MIT talking about how many AI projects fail. Take a look at our score, 94% AI deployment success.
What does that translate into? It translates into phenomenal, deep relationships with the biggest and the best companies. Take a look at what we've got going on. 10 out of 10 on insurance, 8 out of 10 on banking, the top healthcare providers. This really becomes our reference client set. When these guys, the executives here move from one company to another, when they are reached out to by their peers to talk about EXL, you get that thumbs up, and this really becomes our strongest pull. I mean, the logos here are the best of the best, and these are built on years of relationships working with these clients and kind of building that reputation for EXL. Let's go a little bit deeper into, okay, where do we play and what is it that we do for these clients? Now, Rohit alluded to this as well.
The work that we do across these industries is in highly complex workflows. These are not easy, and they are highly regulated. You have to be very precise in terms of how do you deploy the AI and what is it that we are doing for them, and they're customer-centric. It's really, really focused on what is it that they're doing with their customer. With these needs, why is it that they pick EXL? Number one, it's the trusted execution. It's because what they understand that core to the EXL DNA is customer obsession. We are phenomenally customer-centric, and that trusted relationship of saying, "These are the guys that are going to deliver for me," is one of the first reasons that that conversation opens up.
Once you get that opportunity, the execution of it really comes back to the ability to bring that data context and AI together. The context that we built up over years of operation with these companies about understanding that ontology, understanding how is it that they work together, how the data pieces come together, how do judgments get made, what are the guardrails. Finally, the last piece of it, the ability to bring EXL proprietary IP solutions. The ability to start bringing in our IP to say, "Here's an AI that I have built out, and here's something that can get deployed to drive that outcome." The combination of those factors is why EXL gets picked, and it's something that we kind of keep building on, keep delivering, and keep burnishing our reputation.
I'm gonna go a little bit deeper into these examples, and I'm gonna go back to the themes that Vikas talked to you about. Our themes right now, and this is what's really driving the really high levels of data and AI demand for EXL. Number one, we get called in when a customer says, "I need to start bringing in AI solutions. My data is not ready.
Come make my data ready." It's probably the first gate check on saying, "I need help and I need someone to fix it for me." Two, is customers that are talking about saying, "Okay, I need help with both getting my data ready, but I also want to understand how do I reimagine my workflow and bring in some AI capabilities to make that workflow better." The third archetype is one of our existing operations clients who comes in and says, "Okay, I've been running this business with you, EXL. How can we make it better together, and how can we share in the benefits of it?" Right?
The fourth is where it all comes together for us because here's where we've been able to take our data, the context, the AI, our technology, and bring it all together into an industry solution that we run, and we run on an outcome basis. It's the ability to actually say, "I'm gonna be putting all of this together for you, Mr. or Ms. Client, and here's what you're gonna get as a benefit." It's the highest bar, and it's one that we're increasingly moving towards. I'll get you examples of each one of them, and they're gonna start coming to life. Let's start off with the first one, which is, how do we help our customers get their data ready for AI? The example here is a top 20 global insurer.
These are guys, they're in commercial insurance, and what they needed to do was to say, "I need to move to a model where I'm taking my underwriting and my claims processes and making them AI first." Problem's easy. Other, you know, lots of companies have taken a stab at it. The challenge was that the data wasn't ready for it at all. The data was all over the globe in different places, different products, different lines of businesses. In fact, they couldn't even agree on what a definition of premium was. Premium was defined differently across all the instances. What we ended up doing for them is using our capabilities in the EXLdata.ai to build out data pipelines for them.
The ability to ingest this data from these various sources, the ability to parse it, to validate it, and then start feeding it into a new data lake that they were then going to use for driving their downstream work. The benefit for them is pretty clear. What we were able to do is massively reduce the amount of time and the effort that it took for them to build these data pipelines. What ultimately that did for them is once they created the underwriting algorithm with that new data, they were able to massively reduce the cycle time for it. The benefit for EXL is probably even bigger.
Because what we were able to do was take a relationship with this client that was 10 years old, but was largely with the operations side, and expand it to doing work for the CIO/CDO. We took a relationship that was old, that was mature, and we more than tripled it because of the new work that we're now doing with the CIO side of the business. When I talked to the CIO to say, "Why did you choose us?" His answer was very simple: "You guys were the best at bringing together the engineering skill set with the domain knowledge. You guys knew insurance, and you guys knew the data engineering.
That's the value prop that really works for us." That same value prop, because of the work that we've done here, that value prop is something that we're now taking to all of the other insurers, be they commercial, be they personal lines, be they L&A. That value prop is now really working in terms of just the momentum that we've been able to create in our data management business with this particular use case. As you can see, it has like a long tail for us in terms of what we've been able to drive. Let's go to the next case. This one's actually one that really excites us.
Rohit talked to you about how is it that we think, you know, the mid-sized market is gonna become an ideal use case for EXL's capabilities in terms of bringing together data and AI. This is a really interesting case where a mid-sized client came to us, they were in an RFP mode, they came to us and said, "Okay, what is it that you can help us with? We think our processes are fragmented, we think, you know, we need to basically take cost out of those processes." In the old world, this would have gone down the RFP route. They would have basically said, "Okay, let's talk to different vendors. We're gonna select a vendor, we're gonna give some of the outsourced work to a vendor." It'd be a small piece.
We'd probably end up doing a little bit of work for them on claims, a little bit of work for them on underwriting, probably do it offshore. What we ended up doing here, though, was something dramatically different. What we did was said, "Why don't we just design how we are gonna have an AI-first transformation for your work processes? We won't just stop at designing the AI-first transformation map for you. We'll go back to your data assets, and we'll tell you what you need to do with your data to make that data AI-ready, and we'll put it all together for you, and we'll design it." As a consequence, what we've been able to do, is as for the client side, we've been able to create a massive bottom line impact for them.
This is work that's still undergoing, but a massive bottom line impact for them because we've taken the scope really high. Instead of being at a tiny piece that we were doing on outsourcing, which would have probably created $2 million for them, we've made it now something that's really material to the CEO of the business. We've been able to improve their success rate in terms of grabbing new business as well. This is a virtuous cycle for them. The benefit for EXL is phenomenal because what we've done is taken something that was going to be a small outsourcing piece, and now it's become part of a multi-year program to transform the business for them and then to operate that transformed business.
We are gonna be running their DataOps going forward, we are gonna be running the AI-first ops for them. What it's given us is the playbook. This playbook is now something that we're gonna take to all of our mid-sized customers across businesses, across verticals, and that's the playbook that we are gonna be able to say, "Choose us because we can actually help you through our transform operate framework. We can help you really AI-enable your business, all aspects of it, and it's something that we can continue to run." We are really excited about what this can mean in terms of the potential upside. Let's keep going. I talked to you about, you know, business where we were already doing operations work and how do we bring AI into that existing business.
One of the biggest things that everyone talks about is the cannibalization of revenue. It's what's going on with work that you're doing on the CX side, on call centers, does that all go away? Well, first of all, I should point out, and going back to the Goldilocks comment, EXL has a really small component of work that we do on CX. We've never been built as a CX company. Even with the tiny amount of work that we do on CX, we've actually managed to create a tailwind out of it because what we've been doing is actually winning on CX modernization. Here's an example of how we made that happen. The example is for a large U.K. retailer where we were doing some work with them, and it was manual, it was mostly human-centric.
It was a simple conversation about saying, "What is it that you can do in terms of deploying AI to make, you know, to bring in the improvements?" We deployed our proprietary Smart Agent Assist for them. We were running it on our own platform. What they decided was it was working so well for them in terms of the output, because it was I'm sorry. I am skipping ahead, I think. It's one remote control with two buttons. You would think I would manage it. There we are. Success. We, what we were able to do with them, is I don't think it's me.
It's right.
Thank you. Thanks for the assist. There you go.
Yes.
What we were able to do, by deploying our Smart Agent Assist is dramatically increase agent productivity. What this does is actually gives every agent real-time nudges in terms of what should they be talking about to the customer next. What is it that they need to be able to do to resolve that query in the most efficient manner? Clearly a big improvement. What really was surprising to the client was they had multiple vendors at this point. They saw how our AI was performing, and they saw what we were able to deliver in terms of the productivity improvements. EXL's AI then got adopted by our client across the board.
Now they, our client, has a end-to-end AI cycle that has EXL's AI in front and that has, if, you know, all of their scope, internal, external, everything kind of following it. Here's the interesting part for us. As you can imagine, when we took that human work, we embedded the AI, there was a little bit of revenue that dropped because, you know, that work is now the number of hours that you're spending is fewer for the same number of calls. Our revenue for this client actually went up 20%. The reason it went up 20% is now we are responsible for a larger scope of work, now we are responsible for more calls, and it's our AI that's actually embedded within everything that they're doing.
Think about it. We delivered savings back to the client, but because of that scope expansion, we are up 20% and our margin's gone up because now we've taken away T&M work and replaced it with AI. It's a really powerful use case for us. I will now know what to do. There you go. Let's move on to the last 2 examples. These are both examples where EXL owns the entire platform. These are run on EXL platform where we are providing a service to a client on an outcome-based model. The way it works is all the data is, comes to us. We own the AI, we own the model improvements, we own the execution, and most importantly, we own the feedback loop, which allows a model to keep getting continuously better.
The first one I wanted to talk to you is about collections. Collections is something our analytics business has done for years. We've solved every single collections problem that there is. We've done it for the large banks, the mid-sized, you name it. The collections industry had a huge problem, which was nobody would pick up landlines anymore. The way you have to do collections now is through text. The way you have to do collections is through basically reaching people out in different ways. What we've did is built an end-to-end platform that married the analytics, that married the know-how in terms of which risk tiers to contact, what should be the treatment for each risk tier. We married that with the pipes, the digital pipes of saying, "How is it that I'm gonna reach out to someone via text message?
How do I make that equation work?" It was really, really powerful because it brought together two discrete elements within the industry and put it together into one platform. What that's done for our client is actually a phenomenal lift in terms of the customer contact rates. Customers are now getting the text on the phone, they're able to respond very clearly. Because we are still marrying it with all the analytics on who to, you know, contact, how to contact, what kind of a structured plan to offer them, it's actually led to a massive reduction in the charge-offs. Now, put this in context. A 20% charge-off reduction is huge when you think about the scale of some of these players. It's a, it's a huge benefit for them.
As a consequence, we've been selected as the global collection transformation partner. We are now getting a huge amount of scope going through our system. Here's where the value really gets even better for EXL. Our collections platform now has so much volume going through it, so much data, we've been able to now just keep improving the algorithm, keep improving the collections efficiency, and today we are with 20-plus customers, 20-plus clients. Let's play this forward. We know that the credit environment in the U.S. is gonna get worse. You're already seeing that on the private credit side. As that credit environment deteriorates, we expect to start seeing much more volume churn through our system, and we expect to kind of keep adding to the value that we are providing to our customers.
This is to us, you know, it's really the start of something that is going to be a really strong integrated solution. Let me end with the last example, and I probably saved the best for last, which is EXL's Payment Integrity solution. As you know, with Payment Integrity, we are a market leader. We work with four out of the top five national payers in the U.S., and we're churning an enormous amount of claims volume through our Payment Integrity system. Last year, we identified $3.2 billion of claims for our clients.
I'm gonna talk to you about a use case which is an example for one of our largest clients who actually came to us and said, "Look, EXL, we know you're doing very well with the collections. What we want to do is try and reduce the amount of time and effort that we're putting on collections that are post-pay. Post-pay, just as a quick explanation, is I've already paid the provider or the hospital, and now I identify something wrong with the claim, so I'm trying to reclaim some money back. As you can imagine, a lot of friction in that conversation. Nobody likes to give money back. There's a lot of work involved in terms of the back and forth.
Wouldn't it be nice if instead of paying them an incorrect amount, I found that error upfront and never paid them that incorrect amount? That's what prepay is. What they wanted us to do is to say, "Take all of your logic, bring it to before the provider gets paid, and fix the error before it happens." Easier said than done because if you just move it there, then what happens if, you know, you're not efficient enough at catching the errors? That's where I think EXL really outperformed. What we've been able to do is take our algorithms, modify them, bring them up into the prepay cycle, and maintain that efficacy of the models. What we are in certain categories, we're performing about 50% better than our competition when it comes to the categories that we are looking at.
That outperformance, married with the fact that now it's before the payment gets made, has really been a huge unlock for our client. What we've been able to do is actually massively shift their efforts into the prepay side, which reduces the friction in the whole system. Take a look at the scale that we are running at right now. That prepay program, the total program that we run for this client, is delivering more than $600 million of annual saves. The benefit for EXL is really, really clear. We took that account, one of our largest accounts to begin with, and we've doubled it, more than doubled it over the last three years.
Now the increased volumes that we are seeing, both on the prepay side and the post-pay side through our system, through our model, has allowed us to actually keep strengthening the advantage that EXL has on PI. Now when we go to a client and we talk to them or a prospect and we go talk to them, it's really, really easy for them to see the output that EXL brings to the table, the efficiency of the model, the effectiveness of the outcome, and say, "Yeah, straightforward. I'm gonna move you to the top of the line, and I'm gonna give you more market share." You're seeing some of that kind of flow through in our healthcare results as well. It's really an instance of all of those components coming together and really delivering that outperformance.
Let me end with what does that mean for our overall value creation story? Our target addressable model market is expanding. We keep adding to new areas that we are reaching out into. We keep adding to new places where EXL previously wasn't a big player, but now is. Our demand vectors are extremely strong. Our clients keep bringing us in as their data and AI partner of choice, and we remain customer-obsessed as always about those particular areas. Now we have the ability of actually capturing that with all the IP and the solutions that we built. It's the stuff that Andy showed you in terms of the platforms that we built out. It's the vertical AI solutions that we have within PI, within Payment Integrity, and the other tools that we have.
All of that really comes together to create the virtuous cycle for EXL, the sustainable value advantage. The simplest way I think about it is this is why we've won in the past. This is why we had that massive delta against the industry peers, and this is why I believe we'll continue to win into the future. With that, I'm gonna invite Maurizio because at the end of the day, you wanna hear about how all of this translates into EXL's financials. Maurizio, over to you.
All right. Thank you, Vivek. Good morning, everyone. Thank you for coming to Investor and Analyst Day today. It's really a pleasure to see everyone here. You've heard a lot about our discussions around the market, also around our TAM that continues to build over time. You've heard a bit about our advanced capabilities, also how we differentiate in the market, also how we go to market now, really drive value for our clients. I'm gonna try to bring that all together into our financial model, also talk about our performance, historical, then also our momentum going forward, which is really important. A few key messages that I'm gonna focus on in my presentation.
One is our industry-leading performance, and I'll show a bit of a comparison between us and our peers. I'll talk a bit about data and AI. Our pivot to really drive our sustained growth quarter-over-quarter, year-after-year. Then I'll also talk a bit about our strong balance sheet and also our capital allocation really going forward. First off, let's talk a bit about our market-leading growth. When you look at this chart, this is the last 9 quarters of our year-over-year quarterly growth versus our industry peers. You see that we are growing each of these quarters at least 2x versus our peers, right? In every quarter, including the most recent quarter, the first quarter, we grew at almost 14%. Our peers are right around 6% overall.
What's driving this at the end of the day? I'm gonna go back to what Vivek talked about in his slide. On the right side, it said, "Why does EXL win at the end of the day?" It's really those three vectors that we talked about. One is data context and AI. We have been performing extremely well operationally over the years, and that sets us up extremely well to implement and adopt AI into our client workflows. Two is our investment in AI IP solutions over the years in our targeted segments that we operate in. That really helps us continually win quarter after quarter. Lastly, which is really important, is our trusted client relationships with so many industry leaders that we have developed over the years.
You know, our average client tenure is over 10 years, and Vivek just talked about that, right? When you have clients that you have for so many years, you develop deep, trusted relationships, and they trust you. In doing so, we're able to implement AI that much better and continue to grow the business overall. This market-leading growth and our growth rate just overall in terms of our top-line growth really helps us drive the whole financial model overall. If you look at the execution of our financial performance over the last 5 years, it's really driven right off that top-line growth. Rohit talked about it, we want to drive EPS faster than revenues, right?
You see that year after year, including the first quarter, where we grew EPS at 20% overall versus 14% revenue growth. Even more importantly, you know, when you dig into our financial model and you look at the different levers, you continue to see growth over that five-year period. Our gross margin during that period, and I'm gonna talk about that even later in terms of investments, grew 350 basis points during that five-year period. Our adjusted operating margin similarly grew 360 basis points during that five-year period, and our return on invested capital has gone up significantly over that five-year period, well over 1,100 basis points between 2020 and 2025. What's driving our growth overall in our business over the last five years?
If you look at our data and AI-led business, without any revenue from AI-embedded operations, it grew 21% over that 5-year period overall. More importantly, our total operations business overall, that includes data and AI-led operations and non-data and AI-led operations, grew 14% during that period. You see a very healthy mix between both sides of our business. Now, as we embed more data and AI into our operations business, you're seeing that total overall data and AI-led of our total business continue to climb. When we look at total operations overall, you know, in 2025, almost 20% of our revenue and operations was data and AI-led overall, and that's a climb from almost zero back in 2020.
Why is that really important and why we embed AI data and AI into our client operations? Because it becomes IP-led, it becomes extremely sticky with the client, right, overall, and we end up owning much more of the overall workflow going forward, which is really important. On top of all that, Vivek pointed out in one of our examples, and that happens in many different examples, we end up with more revenue on top of that once we embed data and AI, which is really important at the end of the day, and I'll get into another metric later on that really highlights that. That's really important for us to really continue to build the business overall and become that much stickier with our clients.
You're seeing a transformation in our business whereby so much more of our business overall is becoming more data and AI-led. When you add the overall data and AI-led operational revenue on top of our data and AI-led revenue, you get to 55% of our business in 2025 was data and AI-led overall versus 38% back in 2020. That just shows the transformation of our business over that 5-year period. You'll continue to see that because the amount of investments, and I'll talk about that in a little bit, that we're putting to data and AI is significant now going forward, and it's gonna continue to drive our overall business and become much more of a bigger part of our overall revenue base.
You can see that increased data and AI-led present penetration in Q1 overall this year. I just showed you in 2025, 55% of our overall data and AI-led revenue was 55%. In the first quarter of 2026, you saw it increase to 60% overall, which is really important. You're seeing that continually every quarter. Even more importantly, you're also seeing in the first quarter the health of the overall business. Not only did our total data and AI-led business, without any AI-embedded operations revenue, increase 18%, but you see our total operations business grow 10% overall in the first quarter.
You can continue to see overall, Vikas talked about it a bit why we continue to see very good growth momentum overall in our total operations business, and that will continue going forward. We have a nice balance between the two, and what on top of that, we're seeing much more of our revenue become data and AI-led, and that's really going to help us really transform the company in the years to come. When we look at revenue by industry vertical over the last 5 years, you see us continually drive the overall business in each one of our segments. We've had very good growth in every one of our segments. We don't have a segment that is slow growth or hurting our overall revenue growth or top-line growth overall.
You see insurance growing 14% over the last 5 years. Healthcare, you know, driven by Payment Integrity and a few other areas at 18%, banking, capital markets and diversified growing 20% during that time period. Each one of our segments is growing extremely well and driving the overall top-line growth. Overall, you're seeing us continually get more diversified globally. You know, 14% of our revenue in 2020 was driven by our international segment. It's up over 17% now in 2025, and we do see our international area to be that should be growing at or above the overall growth rate now as we've seen over the past, and that should be the case, you know, going forward also. Let's talk a little bit about our overall business model.
This is a little bit of a reminder and also a new metric that a number of investors have asked us about. First off, you know, over 3/4 of our revenue continues to be recurring. We have a very annuity-like revenue base. When we talk about recurring, we talk about revenue that is contracted 1 year or more overall. We continue to see this. This has been our historical trend, and it continues to be our trend. What we've also seen is our net revenue retention be greater than 1.1, particularly in the first quarter of this year and also in 2025, which means of our current recurring revenue, we continue to increase that overall on a quarter-over-quarter annual basis overall.
That's really important in this new AI era, where, you know, a lot of concerns are about cannibalization and revenue coming down from existing clients. That's not the case in our revenue base. We are embedding data and AI, but what we're seeing is getting a bigger moat within the client and also driving the overall total revenue for that client, and that's reflected in that metric. It's one that we're very proud about and one we'll keep, you know, displaying now going forward because it's an important one. It's also been a concern for investors, you know, that has been brought up a number of times.
I talked a little bit about our gross margin expansion over the last five years, and you can see it grew 350 basis points over the last five years. What that has done has given us the ability to increase our investments more than 4x in the last five years. AI needs investment. If we're going to continually build AI solutions and continue to really work with our clients to embed that into their workflows, it needs investment, not only in the solutions but also in R&D. The way to fund that investment is to drive gross margin, right? We've talked about that a lot in our earnings call, whereby we are looking to drive gross margin, but we're also going to be spending on investments to drive AI solutioning, and also in R&D.
That's what that $81 million now is really comprised of. It's us building AI solutions now going forward and also spending a considerable amount of money now on R&D to really find those solutions, work with clients, whether it's a POC or something that we are building ourselves, to really drive the overall top-line growth. All of that is going to help us continue to drive that percentage of data and AI in our business, in our total top-line growth to 60% and beyond that. Now let's talk a little bit about our capital allocation. You know, and we really have been working on our capital allocation to really drive shareholder value. Our return on invested capital has gone up materially since 2020.
Back in 2020, we were less than 9% in ROIC, and we're well above 20% in 2025. Actually, the first quarter was even higher than that. What you're seeing is a meaningful increase in our overall return on invested capital. How have we done this over the last five years? It's really driven by two things. One, it's us increasing profitability overall over the last five years. You saw the increase in AOPM during the last five years and the increase in EPS. That's helped us drive ROIC.
Also being disciplined with our overall balance sheet. You know, we've been fairly prudent, I would say, with our balance sheet and our capital allocation over the last 5 years, where we've done a number of tuck-in acquisitions or smaller acquisitions, and also allocated a considerable amount of capital to share repurchase over the years. That's really has helped us drive our overall ROIC. When you look at kind of going forward, the level of capital that we can allocate, you're seeing some very strong metrics. First off, our free cash flow now in 2025 was virtually $300 million during that period, and it's up 34% from the prior year period. Our business is generating a significant amount of free cash flow that gives us the ability to allocate, you know, and deploy going forward.
What you also see in our balance sheet is a fairly modest overall leverage balance on our balance sheet. Our leverage as of the end of the first quarter was less than 1 time overall, which puts us in a great position now going forward to be able to allocate capital. If you know, if you think about a conservative 2 times leverage for the overall business, and you add this, and you add our free cash flow that we generate on an annual basis, that gives us plenty of capital now to really allocate to that 2 big levers going forward. Those 2 levers are M&A overall, and then also allocating capital stock buyback.
What you've seen in the last three, four years is us being a bit more tiered toward allocating capital towards stock buyback, which has done well for us over the years. Now you'll probably see a little bit more of a balanced approach. You know, we talked a bit about the capabilities that we need going forward to really drive top-line growth, particularly in data and AI. That will result in us allocating our capital allocation a bit more balanced between M&A and stock buyback going forward. Let's talk a little bit about our metrics. Here is our 2026 guidance. When we released our first quarter, we increased our guidance from what we started with at the beginning of the year.
We started the year at 9%-11% in terms of revenue growth for the year. In the end of the first quarter, with our outperformance, we increased to 10%-12%. What we do see going forward in our medium-term target, and when we talk about medium-term target, we're talking about 2026 and 2027. We continue to see double-digit year-over-year growth in our business. We have the momentum in our business today with all of the investments that we made. When you look at our pipeline, it's still very robust and very healthy, and that gives us the confidence to tell you that we continue to see this double-digit momentum well into our medium-term target, which is the into 2027 and the end of 2027.
When we think about AOPM, we do think about a fairly flat AOPM for 2026 when you compare it to 2025, but we do see the opportunity for incremental improvement in our AOPM in our medium-term target going into the end of 2027. Lastly, you know, our adjusted EPS guidance today is 12%-14%. That's up from 10%-12% that we started the year at. Again, we had a very good first quarter, that puts us in a great position to increase our guidance for the year. We'll continue to revisit that, by the way, on a quarterly basis now going forward. For us really going forward, you know, we continue to look to drive EPS faster than revenue growth.
You saw that in my prior slide, whereby over the last 5 years, we've virtually done that every year. We continue to focus on that, we do that through an incremental improvement on AOPM, also optimizing everything below AOPM to really drive that now going forward. In summary, you know, Rohit talked about, you know, that we're well-positioned or well suited to thrive in this AI environment. You know, if you look at both sides of our business, both our data and AI and total operations business, they're both symbiotic, they are both growing very healthy now going forward.
We've made a significant amount of investments in AI, particularly in AI IP solutions, that gives us the ability to really sell into the market and really capture more market share now going forward and continue that double-digit revenue growth versus our peers overall. Overall, Vivek talked about our growing TAM and our ability to really succeed in really going to market with an increasing TAM and also those trusted client relationships. Overall, you know, we believe there's a very solid bedrock there for us to really continue to grow double digits now going forward. With that, we will head into Q&A.
Thank you all. I think the way that we're gonna do Q&A is we're gonna get some chairs up on the stage and have the presenters up here, and then we're gonna have some microphones. Because we're being live streamed, please wait for someone to get you the microphone to have before you ask your question so people can hear it on the webcast.
Inside.
Yeah, just why don't you guys just sit, and I'll grab a seat, and then we'll start taking questions.
Yeah.
It's great with that.
Next to it.
Go ahead. Thank you.
First question. Over to David.
Yeah. Hey, guys, David Koning at Baird. Thanks so much for this. It's great, and congrats on good growth. I guess my question is around the cost of AI and the tokens that seems to be ramping. We get a lot of questions on that. Can you buy those in bulk, or do you pay for those? Do your clients pay? Could you buy them in bulk and then resell and take a little profit? Maybe talk about all the, you know, costs and benefits of the tokens.
I could take that. Firstly, you're right. Sort of company reached a certain point, then the reasoning model was introduced as Vikas was talking about in that AI spectrum, the cost suddenly went up more. The more you reason and you're using it just sort of keeps consuming more tokens. Firstly, by design in our solutions and how we are using the harness layer, et cetera, we are always looking at how to manage the token cost better. That's the first lever 'cause in enterprise you can't scale. Second lever, I talked about the deterministic logic use. Wherever you don't need the LLM, you actually don't use the LLM, you use the deterministic logic, which is the other thing as I mentioned when I was presenting.
Thirdly, the other important thing is that you're also making the choices for use the right model for the right context. Not everything needs a Ferrari, so don't do a Ferrari. First big focus is the token economics because without that, enterprise value can't scale. Specifically to your question, absolutely, we have relationships in which we have preferential rates for tokens. We have our own LLMs, which actually, like I mentioned to you, are 1/10 of the cost because they are like a 7 billion parameter model versus a 70 billion or a trillion billion parameter model, so that helps. That's one.
Two, in some instances, what clients will say that, "Look, we will manage this token part of it because we have a broader enterprise arrangement, and for that, we will make sure that we consume that." For all our solutions, we have actually procured that compute space. That's why we also have the partnership with all the hyperscalers, and we have made sure that we do that all the time. Here is the other amazing thing. Every time we keep engineering well and keep using it well, and we save it, if you're charging on value, that sort of accrues back to you. That's sort of how we've been approaching this.
I just wanted to add to that last point that Andy made. We showed you certain outcome-based, you know, client stories. 30%+ of our revenue today is outcome based. When you're charging outcome based, when you've got end-to-end accountability, you absolutely have the ability to actually say, "I'm gonna choose how I'm gonna manage the tokens, but my customer's gonna pay me on outcomes." You have that delta there.
Hi, this is Puneet from JPMorgan. Thanks for doing the presentations. Really helpful. From the presentations, it's clear there are a lot of changes happening in the industry and how you deliver services to clients. The change management will be very important for EXL as well, as much as it's for your clients. Talk to us, how are you measuring your success in executing against your internal change management challenges, making sure that the services that you deliver, that it's really AI-first services rather than just adding AI to people-based services, like at the foot soldier, the folks who are engaging with customers, that they are executing against that agenda.
Makes sense. Yeah, go ahead.
You're right. Moving towards AI is as much a change management challenge as it is a technical challenge, and many times the change management turns out to be more difficult than the technical challenge. I think there are three or four key things we are doing in EXL. One is for the operations business that we are actually running. We have a very clear metric of progressively moving those operations to more AI-enabled and potentially more AI-led. All of our operations leaders actually have this target that they have to move it. Remember, for us to move to categorize an operation as AI-led, not only the AI has to be infused with effectiveness, but the contract needs to be modified 'cause the client needs to acknowledge contractually that it has moved.
That is one clear metric which is driving operations more towards AI-led. As far as the data and AI businesses are concerned, that's a very clear metric. More services, more solutions, and revenue is something that drives us to what kind of growth. You saw the 18% growth that we're actually driving in that business. We are also realizing that we need to drive innovation to support this at multiple levels, so we have three levels of innovation. There is what we call the democratized experimentation, wherein we are inviting all our colleagues to bring in ideas and prototypes and solutions to the table. We run an annual event. For example, last year, in the Idea Tank event that we actually ran, we got 11,000 ideas with prototypes, so that's another metric which is using as to what sort of innovation happening.
Of course, we have ideas progressing from there to R&D and to then funded projects that we're actually doing. There, the metric is all around how much of investment is actually happening in creating those solutions, and Maurizio spoke about the investment they're actually deploying into that. That becomes the second metric.
The third metric is how the talent pivot is happening. You know, given the business that we are actually running, clearly there are two kinds of colleagues. You know, one is people who are working on AI. These are the people who are actually creating AI solutions, the data structures. I spoke about those 17,000 plus people we have, which are data scientists, engineers, AI engineers, architects, and so on, so forth. The whole thing there is creating more capacity and more capability and diversifying that talent pool. The rest of the organization also needs to participate because while they may not be directly working on AI, they are the ones who are actually bringing in the context that I spoke about, as well as have to eventually work with AI.
We call them as colleagues who are working with AI. They also have to go through that pivot. These are the three or four things that we're actually doing. Finally, it's also, you know, treating EXL as the client zero, which is to say how much of AI are we infusing, you know, in our own operations. I can give you an idea, for example, you know, last time we actually had the earnings call, we actually had an AI agent available to Maurizio and Rohit who were basically querying whatever information, and the agent was actually returning based on all the data sources we have within EXL on the specific responses to some of the questions that you guys were asking. Just to give you an idea of that.
I'm just gonna add one thing. Puneet, you also mentioned the customer part. Anytime technology moves too fast, people are always behind. What happens is customers struggle, right? With due respect, because things are changing so fast. For us, when you saw that EXLerate thing, it's not just in the lab. We encourage our teams to actually deploy production use cases and make it real. Unless you're working with a customer, that's like a durable moat because if you can bridge that gap and create that value, Rohit also talked about investments in those skills, call it forward deployed engineers to work with customers, so that you can bridge that gap between how fast technology is evolving, where are you, and then what you need to do to deliver value. I just wanted to highlight that part as well.
Got it. Thank you.
Thank you.
Surinder Thind, Jefferies. As a management team, apologies. As a management team, how concerned are you with the pace of change at this point in terms of the services that you provide and the disruption risks that you face? Like, there's been instances where maybe planning last year for certain services, you fast-forward to this year, and maybe it doesn't make sense to offer those services. What you're doing to kind of stay ahead of the curve is, are you planning on a three-year horizon? Are you looking at certain capabilities that don't exist today, and you're figuring that's where the models are gonna be two years from now? Like, how are you working through all of this?
Surinder, first of all, I have to acknowledge that the space of change in this industry is quite, you know, it's very, very fast. If I were to sit here and tell you that we have a crystal ball that tells you where our revenue is gonna be, you know, by line of business with precision three years from now, I'd, you know, I don't think you'd believe us. What we've done is, we are actually planning for a whole range of scenarios. What we are doing is, in our planning, we are building out specific scenario ranges, especially as you go further out, in terms of adoption rates and what that, you know, what that scenario means for us.
The scenarios are really about enterprise adoption rates of AI and how quickly do enterprises make that shift versus not. The way we built our planning is because of that Goldilocks scenario. We actually win in all of those scenarios, irrespective of what that pace of change is, but the drivers of the growth will tend to vary between our different lines. That's one aspect of how we've looked at it. Your second question was, are there new things that we are planning to do? We absolutely are. That's where I think the strategic planning focus has become that much greater for us. There are new capabilities that are getting formulated as we speak. What is EXL's role in creating that capability? What is EXL's role in kind of delivering that going out into the future?
That's absolutely a core tenet of our planning right now, and we are investing in that, and we're kind of building for it.
I'll just add a couple of things. Number 1, the planning cycle for us, which used to be 3 years and reviewed every year, now it's become quarterly. The planning cycle has actually become a lot faster given how quickly the changes are happening because the environment changes very rapidly. Number 2, our focus on where we are going to be making investments and where we'll be looking at doing acquisitions, that's changed significantly. You know, we need to be investing ahead of time before some of these trends become reality, and our ability to be able to kind of create the right kind of foundation and the right kind of capability set, that becomes really important.
We've increased obviously the magnitude of investment, but we are also being a lot more deliberate about where we are investing so that we can position ourselves for this uncertain and ever-changing world in a rapid way.
Hello. Inci Kaya with IDC. My question is with the whole proliferation of agentic agents, AI agents.
This is gonna result or is already resulting in AI sprawl, which means we need to somehow keep track of them, govern them, identify them, register them. Can you speak a little bit about agent registries and if and how you're approaching that?
Sure. Yeah, I can take that. Firstly, we just didn't have the time to go to the full breadth of the demo. What we've essentially done in our agent studio is firstly, we've made it completely compatible to TRUE, Laser, some of the other new technologies that are coming up, so you can use any foundational technologies. What we've also allowed is, within the company, you have to allow innovation to happen so that everybody can contribute to innovation, right? That's the point Vikas was making.
What's important is that what problem you're solving for and your ability to register those agents and making sure that, and I talked about the governance, the state, the traceability, the auditability, that is super critical because without that, you know, most people will stand up on stage and they'll say, "This year we want 100,000 agents." That's not what the message Rohit will give. His point is where is value, work backwards from value to see where it makes sense. By the way, now there are agentic designs that you don't need to have like 20 agents for something. Part of the problem is people have worked in a manner where they're taking the human work and giving to an agent. In some of the new orchestration design, you should be collapsing workflows. You should not be replicating the old workflows.
A, start from use case. 2, make sure you have the technology that allows you to innovate. 3, register that technology in the right manner and make sure at all times it's governed, it's managed, it's observed, and that you're able to monitor the performance. That's how we've been going about it.
That's an important point because what we don't want to do is to curtail innovation in the organization. We ask our people that, feel free to experiment, but remember two things. One, do it within the guardrails and do it using the standard technology stack which is being available to you. Once you think that you're actually getting to a point which where your experiment can get into production use, it has to go through a compliance check, and it has to be registered in the standard platform so it can be monitored and maintained going forward, right. Put those guardrails of development and ongoing monitoring, but then allow the innovation to happen because you don't want to tell people that, "Don't do it because of these concerns.
Hey, Gage Schwartzman with TD Cowen. Obviously, it feels like there's a new AI event every week that every investor wants to talk about. It feels like the past week now has been talking about the OpenAI and the Anthropic joint ventures that they've been discussing. I believe OpenAI raised about $4 billion, aiming to more directly tackle enterprise AI services. Curious what your thoughts are on that? Does that pose any indirect tailwind to you guys or maybe a headwind in the future? Just curious about what your thoughts are there?
I'll let Vikas and Vivek talk about it first, and then I'll add to it.
One of the, I guess, one of the narratives that we've been hearing over the last many months is that AI is going to become so strong or is already so strong that you do not need any stitching around it, you no need to put any context in it. It's basically just drop-in technology, right? Take the technology, drop it into a workflow, and it'll start working. That's generally what we've been hearing.
What we have been seeing and what we've been working on, and that's where we see the opportunity for us, is that, no, actually it's very complex because you can have the core model, but then to bring the data, the context, the stitching together, creating the solution, deploying the solution, monitoring it, and then the change management around it, all of that requires an immense amount of solutions around the core model, as well as services that need to be provided on a one-time basis and an ongoing basis, right? That's our thinking, and that's where we're seeing the demand. In a way that you actually have these foundational model and frontier model companies creating these services businesses to be able to create this in the enterprise is validating our point of view that you cannot just deploy a model.
You need to do that. That's the first thing, that I think this is validating that for AI to become real in an enterprise core workflow, a lot more needs to happen. Frankly, that is where we bring our expertise. You know, we bring in the domain. The second question is, will we end up competing with them? Yes, we will to some extent because as they start getting into this area, we will compete.
I think we have an advantage, which comes in from years of experience on data, on context, on specific industry domains, which I think is something. Established client relationships, enterprises where I think will actually give us a bit of a head start. I think we should feel confident we should be able to continue with that advantage. Vivek, something you want to add?
I just want to add 2 quick bullet points to that. First of all, we've been talking to you in our presentations about the fact that you need data, context, and AI all together to win. For the most part, OpenAI and Anthropic so far were talking just about AI. What you've heard is, as Vikas pointed out, is that approach doesn't work. You need all three. When it comes to all three, we've really got a massive advantage over them on the data and the context side because that's where the decades of work with those regulated industries comes to fore. I would love to hear someone who walks in and says, "I've just learned about insurance yesterday.
Now let me tell you how your claims ontology needs to be designed. You know, we're gonna win that battle every day.
The second part I wanted to give you was an example. I talked to you about the top 20 commercial insurer, which was where we'd won the data work with the CDO. They actually did a hackathon. They brought in 10 different providers from within the industry, all of whom had experience at that client site, and said, "Okay, let's all design data pipelines together, and let's see who can design that data pipeline the fastest, the best." EXL came number 1. The reason we came number 1 was not because our engineering skills were superior to everyone else, it's because we knew the data, we knew the ontology, we knew the context. I think, you know, yeah, we welcome the entrants. We welcome entrants who are, like, validating our hypothesis. In those industries, in those clients, I think we'll continue to win.
Yep. It's a validation of our hypothesis and frankly, the opportunity for us.
No, I think you guys covered. It's great. You know, we have an internal WhatsApp group for the management team, and Anita posted this, you know, in our WhatsApp group and everybody went two thumbs up.
Big thumbs. Big thumbs
Because this is fantastic. I mean, OpenAI is doing it, Anthropic is doing it, Sierra is doing it, Palantir has forward deployed engineers. Every single, you know, new technology application, if it requires, you know, FDEs, is just validating our business model. By the way, they can either try and hire these FDEs, which are super expensive, and try and build it, which is super tough, or they can partner with us.
With us
and we'll do it. you know, it's a great opportunity for us.
We've discussed it because remember, we bring this from multiple directions. We bring it from-
You need to take a mic.
the operations side, we bring it from the data side, we bring it from the analytics side, the same model side. You need all of these components to come together to build that right context.
Just to add to that point. If you think about Google or AWS or Microsoft, they all had professional services arms because that's the way they serve some of their strategic clients. With OpenAI, Anthropic, as they are going into the market and, you know, their enterprises, you know, reach is increasing, they are realizing they cannot service it. They are going to have a professional services arm like the partners or the other hyperscalers have, but they cannot scale it. As the partners have a ecosystem of partners where we are very much entrenched, I think that opens up that further opportunity as you partner with Anthropic and OpenAI, they are using the same model. They are not going to serve 10,000 customers themselves. They are just doing what hyperscalers have already done.
In some ways, it's a huge opportunity for us in terms of how we partner with them, and the realization that it can be done is the validation of our whole thesis.
Okay. Hi. Amar from Everest Group. Great presentation. We have seen outcome-based pricing model tried in the past several times and they have really failed to scale as expected. We talked about outcome-based pricing model today. How are you looking to approach it differently this time so that you are able to scale it and execute it well?
Can I go ahead? First of all, we are already at scale. The outcome-based component for EXL today is more than 30% of our revenue. When you put that into perspective, right? That's already a massive amount of business that's flowing through those outcome-based models. Now, what we've done is we've got a two-pronged approach. One is in industry areas where the client is already used to an outcome-based kind of an approach. We've really gone forward, and we're driving further ahead and increasing our volume through there. Payment Integrity is an example, collections is an example. The industries are already used to outcome-based pricing. What we are doing is we are just grabbing more market share, and we're kind of, you know, building a better product and winning more.
The second part of it is taking our existing business where the client wasn't really outsourcing on an outcome-based way and saying, "Okay, we are gonna take on more end-to-end accountability, and by taking on more end-to-end accountability, we are gonna kind of flip this into an outcome-based model, and there's gonna be a win-win." We are actually seeing some very nice traction on that side as well. The traction on that side is actually coming from AI. Rohit talked about the mid-sized market for us. If you're a mid-sized company and you have to compete with the large guys who already have AI, you have to kind of adopt it very quickly. You don't have the means of being able to do it internally because you don't have the size and the scale.
You want to go to someone who can say, "I have the rails built out, I have the technology built out, I have the AI built out. You don't have to worry about it. Just pay me on results." That's where our wins are coming. I think you're gonna see both of those motions happen. You're gonna see for the large industries like collections and Payment Integrity, we'll continue to drive volume, and increasingly, we are gonna switch to outcome-based for the mid-sized.
I was just gonna add one more thing to what Vivek said. Where the client can't define value or the outcome and the value can't be attributed to us, then we won't do it. Otherwise, what's gonna happen is everything you do, you're gonna pass on, right? 2, it's gonna be difficult to baseline and assess. To Vivek's point, we've been actually very careful. Look at subrogation, look at Payment Integrity, look at claims. Companies are used to, over the years, to send those data feeds out and get the value back. We can control that entire stack wherever it's definable, and you can attribute value, and you can charge on value. We've been very prudent. Places it makes sense, sign up for it. Places it doesn't make sense, don't sign up for it.
Dan?
Thank you. Lars Goransson with IDC. Very curious if you could expand a little bit about your growth strategy on complex domain-specific industry problems versus your international expansion. Or would you consider potentially expanding into other industries that exhibit similar characteristics to the type of problems you're solving in BFSI and health?
Well, first of all, we have one of our IMUs, which is called banking and diversified industries, right? The intent was that we do wanna use that to expand into newer areas, and do that in North America, and then as part of our international growth markets, do it internationally as well. We do think of that as an area where we can start expanding. The areas that we are looking at expanding into, across the board, we are very focused on the large technology players. The tech space is something that is of interest to us. Telecom and communications is one where we've already built up a pretty substantial presence. We want to expand that further. Mobility is an area that we're looking at very closely.
I think if you think about us, there is a plan and a strategy for going after more of the high-tech players and the digital native players that have a very strong spend profile, and that's an area we are looking at, doing it both North America as well as international.
I think we'll take our last one over here.
Hi, team. Chandak Biswas this side, from Avendus Group. In line with Rohit's keynote, I think the noise around AI is also creating an AI fatigue in the industry. We have seen a large number of clients with failures as well in last couple of years, either through internal experimentation or through other incumbents, right? What's your strategy of engaging with those clients, and what's EXL's way to differentiate yourself from others when you engage with such clients with a bad taste in their mouth?
I can take that. Remember, we spoke about today that enterprises are actually moving from experimentation to production grade, and you can't get into production grade all over the place. What means by production grade is that you select a few, but core business workflows and make agentic AI real in that. Our approach with our clients is first to work with them to identify what those areas are, what makes the biggest impact, where the technology maturity is the point where meaningful value can be created, how we can actually build this whole concept of data context and AI to create a solution, and then how do we deploy and drive user adoption in that. It's taken them through this journey. The conversations have changed into this production grade AI deployment with our clients.
They do need help in that, right? Because experimentation is a different thing, but then taking it to actually deploying it at scale and making it effective and outcome actually is very different. You're right. The question that where is the return and when do I start seeing it in my customer metrics or in my P&L, I think that conversation is very real. It wasn't happening about six months ago because it was still, you know, it was a lot around data management, but it wasn't so much about agentic AI, but now the conversation is happening a lot.
I'll just add a couple things to what Vikas said. One, I think you're exactly right. Good advice in this time is priceless because there is so much fatigue, clients are getting things left, right, and center. There is one very interesting metric we measure, how many agentic AI conversations you actually say no to. 58% of the conversation we'll advise the client, "Don't pursue this.
This is not the right one." That's the reason we have 93% success rate, 'cause you have to say no to a lot of things because, like I said, people look at that five layers of cake and they think, "Let me just put this in the enterprise and it works." Secondly, demonstrable evidence where you've done the work and you've had those one-on-one battle combats, you've learned from it, and you can bring it back. That sort of really is valuable for clients. Third, like Rohit mentioned, there's also a significant investment in that talent to be able to have that conversation 'cause it's not just that you're advising them what you can bring to the table, you're also advising them what not to do and how to go about it. Just want to add that context to what Vikas said.
All right. Last one. Thank you.
Thank you so much.
presentation. Great conversation. I'm Sudarshana Bhattacharya from Gartner. I have 2 questions. You talked about human on the loop, right? Moving from human in the loop to human on the loop. How do you see industry reacting to that position, especially FSIs? How do you see the diffusion of roles happening on the end user side, like the CISOs, CIOs? How is that happening? That would, I'll be interested to hear that.
The second one, you mentioned about 40% reduction in effort for data pipeline. If you can add a little more color to that, to what type of use cases where you are seeing 40% reduction? Does it include both unstructured and structured data? Especially, or, and also how are you tackling the fragmentation that's already there?
Yeah.
Again, focused with FSIs.
Yeah, I'll take that and Vikas and Vivek can chime in as well. To your first question Sorry, what was your first question?
Human on the loop.
Human on the loop. Yeah, sorry. That's what They're just thinking about a second question. On your first question, see, that's sometimes terminologies and how people use them. If you saw in the demo, I think the pace at which agents process information is much faster. If you're gonna put human at every step, the problem is you're gonna create more bottlenecks, right? It doesn't mean that every important decision, every aspect of governance, every audit trail, everything where you need that traceability, you can't lose on that. If I just take you back to demo quickly, one of the things that we are very careful of is that let some of the things be informed to an agent to make that decision, human agent I'm talking about, and they don't have to intervene at every step. They're observing what's going on.
In that example, you were able to see that while I got the approval to go ahead with this, but I'm getting a sense from the context that this is something that may be an opportunity for recovery, which otherwise was not. It's still human, but the profile of the human changes. Think about the old days, you had work, divided like this complexity, this complexity, this talent. Now that humans' role is elevated, and they're making judgment about things. Technology is making things available. Human is still making the judgment versus getting everywhere in the process. They're just sitting on the loop and observing this end to end. Don't take this as you're just going to completely ignore the traceability, the evidence, the governance and decision traces, and then you'll make the decision, you're pressing the approve button. That's sort of the first part.
Let me just illustrate that with an example. We talked to you about claims today, right? Let's talk about how claims was dealt, how we would have done AI for AI-led claims earlier. There would be a claims adjuster. That claims adjuster would basically get a claims form. The AI was just focused on extraction. That's it. We were just focused on extraction. We were bringing the information out. The information was still getting served to the claims adjuster. The claims adjuster was saying, "Okay, all that manual work of kind of taking the work out, that's been saved." Right? The saving was more manual, but the role of the person was still making the judgment. Today, in an AI-first model, you're actually trying to say, "No, I want the AI to make the judgment as well in most of the cases.
In some cases, I'm gonna get the adjuster involved." Now for the AI to become good at basically making judgment in all of those cases, you don't really need a claims adjuster sitting on every transaction, but you still need some element of RLHF, reinforcement learning. That reinforcement learning from a human is now a different type of a human, and that is now human kind of working on the loop rather than in the loop. The output here is that the AI accuracy goes up, the number of cases that the AI is solving for goes up dramatically, and your overall cost and the cycle time into this process comes down dramatically. We think this is how AI-first operations is gonna play out.
There's gonna be a huge amount of work for the RLHF human factor in there, and there's going to be some amount of work still for the expert.
I'll just take the second question also. Last comment on this one. It's iteration, iteration. Sometimes it never lands perfectly. Like Vivek said, I have now confidence on this part. I can hand it over to the human because accuracy is much higher. By the way, our subrogation human machine combination today is performing way better than what we had as a human performance before. Way better. It's 4 percentage point difference that we are observing now. It went through iterations. Never just lands perfectly the first time. Once you master the solution, then you know how to replicate and rinse and repeat it. To your second question, see, this is where, and I think this goes back to the token question as well.
The problem is most companies will advise you when they come with platforms to move the entire data estate because it's driven by consumption. When we run these unstructured data, structured data logs, metadata, go read from COBOL, go read from mainframe, and we've been using Claude Code, you know, because we saw great efficiency, but it's years of working on SaaS. It's year of working on COBOL. When you look at that data estate, these things got built 40, 50, 60 years back. Invariably, you find 30%-40% of those logs, code sets are redundant. They're not even used. They're not even feeding to the downstream pipeline. You want to cache it, you want to keep it in some datasets, fair, but you don't need to migrate and spend on that consumption and bring that data because that's not needed.
For enterprise transformation or the use case you're driving, actually, that has no relevance or value at all. Typically, everybody was approaching that, "Let me take the as a state and take it to the to-be state," because you just did not know what lies in your estate. We just made that whole thing transparent. What that does is all these excessive logs that you didn't need, all those pipelines that were created that you didn't need, downstream 1,000 reports that we're using was feeding from some data source, nobody's using them. We are able to clean up all that log, and that in itself saves 30%-40% effort, but more importantly, downstream consumption because you're not just bringing garbage that you don't need.
Well, thank you everyone. It was a terrific day. Again, I'm gonna remind everyone that the demonstrations are in the next room, next to the lunch, we'd hope that you'd stick around for lunch, be able to speak with the presenters, and we've got lots of EXL management around the space. Thank you so much for joining us. It was a terrific morning. Have a great day.