Welcome, everyone. It's great to see a packed room today. For those of you that don't know me, I'm Kevin Toomey. I'm Head of Investor Relations at FactSet. We have a lot to share with you today, but before we get started, I'd like to share a few things from a legal perspective. First, I want to remind everyone that discussions today will be forward-looking. Such statements are subject to risks and uncertainties, which are discussed in our SEC filings. Additionally, we'll refer to certain non-GAAP financial measures. You can find GAAP to non-GAAP reconciliation tables for those measures on our website at investor.factset.com. On behalf of the entire FactSet leadership team, welcome to FactSet Investor Day 2026.
[Presentation]
Good morning, everyone. It's a pleasure to welcome you all today to Investor Day 2026. I'm excited to take you through our strategy at a time of tremendous technology-led change in financial institutions and financial services. What I know from our clients is that they are keen to embrace AI and capture the extraordinary value from this powerful new technology. At the same time, they can only use AI, their ability to use AI relies on their ability to trust AI. Trust in AI is only going to come when it is high quality, delivers compelling value, and is well-controlled. I believe FactSet is uniquely positioned to offer the world's most trusted financial intelligence platform. We have a long history, over 48 years, of serving clients based on our deep expertise, the high quality of our data and analytics, and exceptional client service.
Our strategy is building on these strong foundational strengths to create a proposition for quality, cost, and control that's better than our competitors. Our mission is to enable our clients to have confidence in scaling AI across their businesses. I believe we have the talent, the know-how, the assets, and the client relationships to do this. Today, I'd like you to take away three key messages. First, we have a differentiated strategy that positions us well to do continued growth and benefit from any AI acceleration in finance. Message two, we've transformed our operating model to execute this strategy, and we have a disciplined execution approach to drive productivity and renew operating leverage. Third, our business model is designed to deliver profitable growth and shareholder value. Today's agenda, we are going to take you through our strategy and demonstrate the rapid progress we've made in transforming our operating model.
After I walk you through the details of our strategy, Kate Stepp, who is our Chief AI Officer, Bob Stolte, our CTO, and Goran Skoko, our Chief Revenue Officer, will show you how our product, operating, and commercial engines are working in lockstep. You will hear directly from clients in video testimonials. Josh Warren, our CFO, will share our financial algorithm for shareholder value creation. We will leave plenty of time for questions with all of you. Over lunch, our product heads, John Costigan and Kristi Karnovsky, are going to introduce and showcase a number of new solutions, show you some exciting demos of our agentic solutions that will bring a lot of our new capabilities to life. Outside, we have a number of our senior FactSet product leaders.
Many of them are in the room at the back right now, but they will be outside manning various kiosks to showcase to you how we are infusing AI in all our solutions, and just to share with you the extent of transformation that is happening across the company. I strongly encourage folks to check out the kiosks and just try out the products. The power of the solutions will hit you yourself. Let's dive right in. We have had a privilege of having a strong foundation of deep client relationships over 48 years that is grounded in a number of core capabilities. It is this strong foundation that gives me confidence that we can capture the generational opportunity ahead of us and execute well.
It is a foundation that is built on the trust that our clients entrust in us, and that is what gives us the ability to do what we are doing today. Since our founders first printed fact sheets and distributed them around the street via bicycle messenger 50 years ago, FactSet has benefited from every technology wave. Since our IPO, as you see on the slide, there has been some tremendous technology-led change in the way clients communicate and consume data and analytics. The big transitions were the internet, mobile, and cloud transitions, and now AI. Through each of these technology waves, FactSet thrived. We adapted our business interface, but what remained permanent and enduring was the quality of our data, the quality of our service, the quality of our embeddedness in clients' workflows. With that strong foundation, we have been able to deliver strong operating income, revenues, and shareholder returns.
Our foundational strengths cut across our broad distribution, our connected data, our embedded workflows, and our unique and excellent service model. Our client franchise is exceptional. We serve clients around the world in a variety of different industries in a very diversified way. We serve asset managers, both traditional and alternative. We serve a whole range of market infrastructure participants. We support wealth managers, we support banks, and we support asset owners and corporates. We have over 9,000 clients around the world in over 80 countries. We take pride in serving the leaders in each of our business segments. The top 100 clients account for about 40% of our ASV today, and we continue to grow with every one of them.
Clients benefit from us the more they use us, especially the embedded nature of our products and solutions means as clients open up new offices, expand their client base, add new employees, and just grow their capabilities, they come to us for more and more services. In other words, our business functionality compounds, and this is a great platform on which ASV compounds, driving long-term lifetime customer value. The connected data that we have allows us to operate in a large addressable market of $40 billion. This market is growing at 6% a year, and the reason it continues to grow is because of the secular forces affecting our end markets that are dynamic and ever-evolving.
Some of the largest trends that we see today that are driving demand for our services include the growing asset allocation to private capital, even including high net worth and retail portfolios, the continued innovation in financial products, both the nature and structure of products, as well as their distribution mechanisms. Think custom indexing, active ETFs, tax-optimized separately managed accounts, tokenized assets, you go on and on. Each and every one of these product introductions creates demand for our analytics. It creates demand for our risk analyses. It also places a demand on our modeling, our reporting, and analytical infrastructure, and distribution infrastructure. You can continue with various other trends, but what allows us to support clients through those seismic shifts is the investments we've made in our data.
This example is a great example of how the investments we've made in recent years in a variety of our datasets, some of our most crucial datasets, is starting to pay off. We invested in what we call deep sector data, which is really insights at the next level of detail in a number of different industry sectors. Banks, insurers, metals and mining, industrials. There's a long list there, and we've been investing in it for years, and you can see the growth rates that are coming out of that. Likewise, we now stream data from over 300 venues around the world in our real-time pricing and reference data. This is exceptionally difficult to pull off, and we've built it in a cloud-native way. Just last week, we had our peak volume day when there was high market volatility after the U.S. midterm elections.
We hit a new level of 55 billion Level 1 real-time ticks in that day, and our cloud-native infrastructure just handled the volume spikes seamlessly, and we were able to provide smooth support for clients throughout the day with no issues. I believe this growth is coming because clients value the differentiated quality of our data, and we are able to do this at scale because we have proprietary know-how. Of course, we have unique datasets as well, but our true proprietary advantage is how we do this at scale. It's because we ensure that the data is cohesive, complete, reliable, accurate, and connected, and that all comes from our data concordance capability, our entity resolution capability. We leverage these proprietary advantages, and we do this at scale, doing millions of database updates literally every day.
We have over 50 million CUSIP identifiers in the market today, and the entire financial industry relies on that for the quality of due diligence and curation of the data. This is the connected data foundation on which our strategy is taking shape. Many of you think of FactSet as a screen on your desk, and that it's a great screen, I love it, but what really matters is what's happening underneath it, and that's really not very visible to each of you. This slide is an example just to show you a few different processes across asset management, banking, and wealth, and show you ways in which we provide capabilities to our clients that are deeply embedded in business processes that really, really matter to them. We are not a portal.
A lot of our capabilities are connected into the many of the decision engines that run global finance. Every day, you see billions of calls, data calls, analytical calls on us from bankers who are running valuations on companies, from portfolio managers who are reviewing the size and shape of their portfolios, from risk managers who are trying to understand the active risk in their books, and from wealth advisors who are trying to inform and advise their clients. This is mission-critical work. To just illustrate this, I'm going to pull up this example of a portfolio analysis tool. This is one of the crown jewels of FactSet. It's been built over decades, and it helps you understand the size and shape of a portfolio.
What you see here is a simulation we did last week of a large- cap equity fund, just to see what it was going through in the context of the volatility in the markets last week. Here, you can see what is the performance of the portfolio, what are the sector weights, what are the worst performing industries, what's the active risk. If I double-click on the top right corner up here, you can see that there are 11 risks that are really explaining what's happening up there. What I get more excited about is what's happening underneath the hood that allows us to deliver those 11 risk factors. That's hundreds of millions of calculations.
Across each of these factors, we need to run thousands of simulations to understand, based on the parameters that have been set, looking through cycles, equity cycles, and credit cycles over the last decades. All of that goes in the background into coming up with the actual performance of a portfolio against different factors. Most importantly, every single data point in that entire analytical stream has to be auditable, verifiable, and it is what our clients rely upon when they're making presentations to committees, to regulators, and to their end clients. This is the embeddedness of the work we do. It is something very hard to disrupt. We believe it is our crown jewel, and we are building around it. Let me give you a quick summary of our service model and why it's so important to us.
You've seen the quality of our data and the intricacy of our calculations. To deliver that at scale and to ensure that our clients are getting it in the formats that they require to meet their standards is very complex, and it can be overwhelming for even the most sophisticated client. That is why we built a high-quality service engine. It has three dimensions to it. We have consultants who are specialists, we have analytical and implementation specialists, and we have a managed services team. These three compound, and they together deliver incredible value every day to clients. We have consultants who are trained in the workflows and the context and the domains of our asset managers, our banks, our wealth manager clients. They understand deeply the specific unique client context, including at the individual level, and they are in 19 countries spread out around the world supporting these clients.
They help them with implementations of new products and solutions. They guide their transitions from other providers onto FactSet. They help with user personalization, and all of this drives engagement and adoption and a high customer satisfaction score. Equally, when it comes to large asset owners, asset services, insurers, and other asset managers that have large data transformation work to be done to look at their internal capabilities, we offer them managed services, which is effectively an extension of the internal team, highly technology-enabled, with deep subject matter expertise overlaid on top. We effectively give them operating leverage in scaling their business. That is the managed services model, and it is high margin for us because it is built on technology, and now we are adding more with agents. So this service model is unique, and it compounds the other foundational advantages that we have as FactSet.
Together, these foundational strengths are being recognized by clients. I believe they recognize and reward us for it. We see over 95% ASV retention across the entire book. We also see that the average client is with us for over 16 years. That is trust in us. It has been built up over time, and we take it very seriously, and we believe it gives us the foundation to pursue the generational opportunity ahead of us. Across hundreds of meetings that I have had in the last year with clients around the world, clients tell me that they are keen to embrace AI. And as I said, they are keen to embrace AI they can trust. They have challenges in activating their own internal data, they have challenges in modernizing their tech stack, and they have challenges in scaling agents safely inside regulated environments.
Addressing these challenges and helping them have confidence in scaling AI is going to be a generational opportunity, and that is FactSet's opportunity. FactSet's opportunity is to take the trust that clients have given us over the years and build high-quality solutions that are at the right value point, that are well-governed and well-controlled. With that, we believe we have a huge opportunity to go after new budgets and target technology budgets, data science budgets, engineering budgets that are well, well beyond the traditional budgets and the traditional addressable market that we have had. That is our generational opportunity as FactSet. Our strategy to capture the opportunity is in three parts, and it reflects our product platform. Now, I will go from bottom to top, and you will see this is a feature of the whole day. It is a feature of who we are as a company. It is not intuitive.
Normally, you'd expect for me to start from the top. But we like to start at the bottom because we're a platform company, and connected data is at the heart of everything. So, part one of our strategy is to take the data that we have, the high-quality, concorded data of FactSet from the market, and mesh it with internal client data. Activate internal client data, mesh it, and create proprietary knowledge, and make it machine-ready context. This, we believe, is the foundational building block for enabling AI to scale at our clients. Second, we have an incredible analytical infrastructure, everything that I just showcased. We're going to take that, and we are adding AI capabilities to it to be able to scale advanced analytics to a whole new set of clients across new functional domains and across new use cases.
We believe this is a huge opportunity to democratize high-quality analytics and make it available at the fingertips of our clients. Third, we've taken a significant transformation of our Workstation experience. We've taken the best of what the Workstation has, which is a lot, and we've refactored it into a new, powerful human agent orchestration platform. You will see a lot of it today. We've taken the core assets that we've had, added new capabilities to create a new platform. The Workstation, as powerful and ubiquitous as it is, we are very well aware that the user interface is rapidly evolving in a world of AI. As we've done in the past, we're embracing this change. We've already done this, and you'll see a lot of it today. We believe the opportunity for a refactored new operating engine that really supports human-directed, agent-executed work is enormous.
In combination across these three layers, we believe the core assets that we had, that come with a lot of client trust, embellished with new AI capabilities, give clients the basis to trust AI and to scale it confidently in their businesses. The strategy that I've just outlined, I believe, gives us a lot of credibility and capability in executing and differentiating on the value propositional elements, the crucial trust elements of quality, cost, and control. Now, I'm going to take you through a slide and go through each of these in turn. Quality in AI only comes when there is correctness, completeness, and relevance in the data input to AI.
FactSet uniquely offers correctness because we have an extensive corpus of the right accurate, timely data that comes because we have invested in it, because of our entity resolution, because of our data refinery. We've spent years perfecting this. So, correctness comes from FactSet. But we don't stop there. We enhance completeness, and completeness is enhanced because of our data concordance. We're able to bring together thousands of other third-party datasets, and we can bring in internal client data as well. And last, relevance really, really matters, and this is where client context matters. So, what I described earlier about our ability to build proprietary client context, that's where the relevance of every single input jumps. With that, you get quality output that is materially better than what can be achieved.
Without our data, we believe even the most powerful large language models will not be able to replicate the quality of data and the quality of output that you get when you use FactSet's data. That's not where we stop, because we see clients, every client that we serve has some sort of horizontal AI, and they leverage data from FactSet through our MCP connectors. Our real benchmark is our platform superior and delivering better value than a client that uses a horizontal AI harness and uses FactSet data connected into it? That's our true benchmark. When we run tests, we are almost regularly, we run it almost at this point every week. We are approximately 10% better on our quality scores than that combination. Why is that?
Why is it that even the most powerful large language model with FactSet data does not come close to our quality? It's because of the domain expertise that we add, the context engineering around the data that we bring, which comes from decades of proprietary knowledge of FactSet. It's because of the unique context we bring. It's because of certain unique datasets that we can bring that we cannot distribute to third parties through MCP because of data rights management. It's a combination of our know-how, the capabilities we have, the domain engineering that we do around the large language model that allows us to deliver this better quality. Most importantly, despite any benchmark, the real value add is in when clients really put this to work in their specific unique use cases.
What we really compare ourselves to is, what is the client's actual task that they're trying to execute by using our AI solution? We check ourselves against, did we achieve the outcome? We work intensively with client feedback every day to refine that. That's very hard to pull off if you don't have the ongoing trust, the ongoing deep connectivity, and real knowledge of the client processes. Let me go to cost. Cost efficiency broadly comes from having the efficiency in use, the optimization of large language models, and ultimately scale economies. We bring efficiency because we have proprietary skills, tools, and ontologies that we've built. That really takes any large language model and guides it in a way that it's far more efficient on token burn.
Later, you'll see an example of how much more dramatically efficient it is when an agent uses FactSet's ontology versus not. It's absolutely orders of magnitude better. That efficiency is not enough. The optimization of the entire harness is really important, and that's where we excel. We take all of the traditional tools and capabilities that have powered the FactSet Workstation, the engines, the triggers, the decisioning that we've done through deterministic software, all of these assets are available to us. We leverage all of it in the context that we engineer around the large language models. That allows us to preserve quality and improve quality even as we swap out models.
The overall cost efficiency in general that we look across the whole solution, again, when we look at it across the tasks that clients perform, is of the order of about 30% better than when you use our data through a large language model frontier harness. Intrinsically, cost and quality are linked, and we understand that equation, and we are constantly optimizing. The important thing is, we are saving clients the significant cost burden of staffing high-quality internal expert teams of data scientists and engineers. That becomes very, very expensive. Those teams have to constantly be tuning every time there is a new model release. I believe there has been about 24 model releases in this calendar year, so roughly a release every other week.
The tuning burden when a new model comes out is very significant because you end up with this, what is referred to last night as the jagged performance of Fable 5.2. That is what everybody has to deal with, which is some things get better, a lot of other things get worse, and you have to deal with that. Our clients are not in the business of just experimentation. So if you have to put this to work in regulated processes, somebody needs to abstract away that jagged performance of a new model release. That is FactSet. We provide that capability. We are constantly at the cutting edge, and we are giving capability to clients to reliably use AI. Let me finally go to our control. Clients in finance, in particular, cannot compromise on control. In regulated institutions, mistakes can be catastrophic.
Our long history that I talked about of a trust-based relationship with clients is what we are building off of. We have a strong control posture on AI. The way our posture works is because we have investments in security. Our investments in security are focused on encryption of data, entitlements and controls, and also on agentic cyber defense. Agentic cyber defense is important. When you are dealing with machine speed attacks, you need machine speed defense. We also focus on governance. Model independence is really important, so you do not get locked into one type of model. We also provide agent runtime controls. You need to ensure that your agents will not do anomalous things in the wild. Last, we offer clients sovereignty.
Sovereignty is at different levels, data sovereignty at a jurisdictional level, but more importantly, we want to assure clients that their data is not being trained on, and two, we want to ensure that their intellectual property is being protected. For this, we have made significant investments in technology infrastructure that brings together single tenant as well as multi-tenant infrastructure on the cloud and on-prem. Clients can pick and choose how they prefer us to make those implementations at different price points, but it gives them real comfort that their data is not getting exfiltrated. This combination of quality, cost, and control, we believe, sets us apart in the market. It allows us to deliver a trusted value proposition and allow our clients to scale. Let me give you a bit more insight on how we are able to do this. We built what we call FactSet Intelligence.
It's a new stack that's been built over the last year. We took the core assets of what FactSet has built over decades, and we added new AI capabilities to it. It's a deep stack, and it's really hard to build, and we believe it gives us significant competitive advantage. I'm going to again take you through it in detail from the bottom. The first layer is connected data. I talked to you about some of the core assets we've had. If I summarize them in a nutshell, thousands of datasets that we either source ourselves and/or integrate and deliver, entity resolution, millions of identifiers, and the ability to ingest data at scale. These were our core assets that were brought together to life based on our unique capabilities in data concordance and our ability to do this at scale. We've then augmented this with new capabilities.
Semantic enrichment, which you see towards the corner here, ontologies, which I referred to briefly, vectorization, and all of this comes together in making the data machine-ready, and it's now resulting in a growing and expansive library of MCP endpoints. Not only are we making this available from the market to clients, we're making all of these capabilities available to our clients so they can activate their data and put it behind MCP. That's a game changer. Likewise, we have a very robust infrastructure for analytical modeling. We have backtesting environments, we have model libraries, we have optimizers and calculation engines, and we have an ability to generate reports at scale and vault our analytics at scale. These are fundamental and require even more investment in the future because the volume of analytics is going to explode.
What we've done with that is we've added coding agents, and we've invested in the infrastructure. We've now built hybrid infrastructure where clients can bring all of this capability closer to where their data resides and their modeling environments reside. Here, we are talking enormous volumes of data, like orders of magnitude more than the typical data we all work with, because quant teams work with, I would say, literally thousands and millions times data that everybody else works with. That requires Spark environments, new clusters that are working at much, much higher speed, and we've built that hybrid infrastructure, and we're having great success, and you'll see it outside. Last, we've taken the best of the Workstation's components. The Workstation is, again, I think of it as a container.
What's really powerful about the container is all the components that sit underneath the surface, like an iceberg beneath the water level. You have calculators, you have optimizers, you have screening engines, you have entitlement engines, you have authentication, observability. There's a lot of core assets that we started with. However, this is the layer where we had to add the most new capability to prepare ourselves for a human agent new orchestration platform. We've added a skills library. We've got an agent-building platform that's building agents at scale. We've now got optimization engines. We have a smart router that allows us to route workflows to different models at the right time. We have a gateway that allows us to prevent anomalous activity at the gateway level.
We have runtime controls that gives our agents a map of an allowable action graph of what they are allowed to do, so they do not run away and create anomalous activities. Most importantly and powerfully, we have a memory engine. A memory engine is the core of any recursive learning with AI. That is what will allow our AI solutions to learn from their end user and make it even more personalized in the future. All of this is coming together, and we have a UX generator that is going to be dynamically rendering UX in context based on the memory, based on the preference of every single user. All of this has been built. This is really hard to pull off, and we could not have done this if we started from a standing start.
It was possible to do this in a year because of the decades of investment we have done before in all of the core assets that are feeding this stack. Now, we have clients of different types, and today, clients believe they have the choice to build all of this themselves with using the power of AI. We certainly have clients, the most well-resourced institutions, who are building and launching a lot of this themselves. But each and every one of them is our client, and what we hear from them is they are keen to partner with us. Why? Because they see us adding speed and agility and complementarity to what they are trying to do internally. This stack that you see here is not a monolith. Every single component is individually available as a module, and this is exciting to our largest and most sophisticated builder clients.
They like to mix and match, take these components, insert it into their own tech stacks, and accelerate. You will see some examples of how this is happening later. That is really exciting to us. This is what gives us the confidence that we are really well-positioned to be this trusted financial intelligence platform. We have had a philosophy of an open architecture and an API-first approach for decades, but I believe this is finally coming into its own in an AI era. Now, this stack is not a stack in isolation. It is the stack on which every product in FactSet is now being built. This is a growing catalog of AI-native products that you see coming out in the market. Each of these products on this page is live in the market today. At the top, you see the intelligent workflow products.
We have organized these products by the end-user personas, the largest personas that are customers of FactSet, advisors, bankers, and other capital market professionals, and analysts and portfolio managers, and investment professionals. Under each of those, there is a suite of agentic solutions that are out. There are some that are live in the market that you will see today. There are some that are coming soon that you will also see today. But the most exciting part of it is our ability to now build and launch solutions under each of these banners is better than ever before. The stack that I talked you through, FactSet Intelligence, is really powerful, and it speeds up our product development. We have products at every layer, connected data products, we have advanced analytics products, and we have intelligent workflow products. This product engine is starting to be recognized by our clients.
I'll share with you a few examples. I like looking at visuals. It just makes all of this come to life more. This is an example of a product in that connected data layer. We call it Data Navigator, and it's really about discovering our product universe. This is very difficult when you have thousands and thousands of datasets with time series spanning 40 years at a time. It's really hard to interrogate it, understand the full nuances of what we have, and Data Navigator does it beautifully. We've taken the platform that we had, our traditional data catalog, and we've fully modernized it for an AI environment. You can access all of this through MCP, or you can access it now through a programmatic natural language chat. You can just go in and say, tell me what's in your deep sector data.
How long can you go back in history with the mines in South Africa? At what level can I understand the mines? You can just go on and on, and it's all in natural language. We have clients starting to use this, both programmatically and with natural language. Here's an example of FPE, which is FactSet Programmatic Environment. This is our quantitative environment. You'll see it outside. It's part of our Quantitative Intelligence suite of intelligent products. It's fully powered with coding agents now, and you can build really, really sophisticated, very complex analysis in near real time now. Here's what a satisfied user has to say about it.
You can see in the tone of this quote that what really matters is the hybrid infrastructure, the quality of the data, and the ability to mix and match client insights and client data and client analytics with what we bring to bear. We are unique at this, and we believe there is huge opportunity to scale this. Lastly, my favorite, I've been going around already marketing it this morning to folks saying, "Just download your mobile app and start using Ask," because it's our conversational agent. This is what I use as my primary interface to learn what's happening in the world now, because unlike a traditional general purpose horizontal AI, this has context. It knows it's Sanoke. I absolutely trust any question I put in it not to get distilled, and it has quality information coming through it. Source- linked, referenced, high-quality research.
It's got what I actually need when I prepare for a meeting, when I'm meeting one of you, when I'm meeting one of our clients, when I'm trying to keep on top of what is the implication of Fable 5.2, I go to Ask. I strongly encourage each of you to try it out, give us feedback. We are constantly learning and improving. What powers this is the same MCPs, the same MCPs that our clients consume, but with a lot of context engineering on top. That's what brings high quality to this. Now, it's not just enough to have great products. We need clients to pay for it as well, don't we? This is some examples of early client traction that we're seeing. Each of these clients on this page is a leader in their own field.
They are clients who have enormous internal teams, sophisticated data science and engineering teams, AI teams, and they're building and launching agents by the thousands. But with each of them, our solutions have been resonating, one or more aspects of our solution. I'll just share the examples in a little bit more depth. The first one is one of the world's leading asset managers, alternative asset managers. For them, the ability to access more of our data through MCP has been a game changer. It just allows them to make our data available to a lot more personas within them, personas that were historically inaccessible to FactSet. In the middle example, this is a large global wealth manager that is rolling out FPE, what you just saw, to thousands of internal users.
And in the last example in the top, this is a large global investment bank that is working with our FactSet AI for Banking solution, which is a part of our Capital Markets Intelligence suite, and making it their primary platform for human agent work. Goran will talk in a little bit more detail as to how they're using it and the upside that he sees in the way this is getting rolled out. I get even more excited by this graph. This is, just to explain what it is, it is the graph of individual end user usage of our MCP over time, ever since we launched our proprietary MCP server at the end of December last year. As you can see, we've been adding more functionality, opening up more datasets, and improving and fine-tuning our capability, and we'll continue to do that as we always do.
What I love is the curves towards the right side of the slide and how much more steep they are. So, as the capability is building, clients are adopting, they're engaging faster, and they're getting up the curve faster. This is a curve I like to see continue. At this point, at the end of the last quarter, we had about 650 clients that were using our MCP. That's still a very small percentage of our total client base. I believe we are just at the early innings of client adoption of AI. This is happening, this steepening of the curve is happening because they're able to trust the quality, the cost, and the control of the solutions coming out of FactSet. This is what gives me the confidence that we, as FactSet, are sitting on a generational opportunity to build the world's most trusted financial intelligence platform.
Now, executing against this requires a new operating model. It requires us to execute with discipline, because there is a lot of change. It's a new way of doing business. To do this, I've assembled what I believe is a world-class leadership team. We paired leaders of FactSet who built the company with leaders from the market who have led data, finance, and technology across the industry at leading institutions. This team has come together. We are transforming our operating model actively as we speak, and they've made great progress. Through fiscal 2026, just over the last several months, this operating model transformation was well underway. As it was underway, we outperformed. We delivered our strongest ever quarter in the last quarter. Now, the way the transformation is happening is along three operating engines. It's a product engine, an operating engine, and a commercial engine.
These are the three engines that are working together in lockstep and preparing ourselves to capture the value from the strategy. The product engine, we continue to make investments this year in improving our products' foundations, and we are preparing ourselves to capture the upside with the new product roadmap that I just shared. With the operating engine, it's all about reliability and operating leverage. With the commercial engine, it's about deepening client relationships and preparing ourselves for a new commercial model. Let me take each of these in turn. In the product engine, some of the investments we made through the year included making our data machine-ready, which was an important part of that MCP curve. It's also continuing to invest in differentiated content, whether it was our real-time content, our deep sector content, and a whole host of others.
We also made substantive progress on our private capital strategy. We invested in partnerships. We've invested in a variety of operational elements that bring together private capital capabilities, both for investors, for dealmakers, and for wealth advisors. Based on that foundation and the transformation of the financial intelligence stack that I talked through, we now have the ability to deliver AI products at scale across every layer of that stack. The operating engine, you'll hear Bob talk about this in a little bit, but we believe there is a lot of operating leverage that we've gained because we've been focusing on simplifying to scale. We've been focusing on strong controls to prepare ourselves for an agentic world. This is really, really important foundational work. On top of this, I'm excited about the agentic transformation that is still in early days.
We now see about 60% of the new code that we are committing on any given day is being written by AI author, it's authored by AI and agents. That's exciting, but there's still a long road ahead. Similarly, about 30% of client queries that come in to our consultants are being automatically handled by what we call Client Assist, which is our autonomous agentic infrastructure that deals with really complex, sophisticated questions from our clients, and we use a whole suite of models to support that. Lastly, we are seeing real progress in our data. So we have a data infrastructure, as you can imagine, with thousands of our experts collecting, curating, and distributing, and developing data products that are distributed. When we modernize one of those data pipelines, we are seeing extraordinary throughput improvement.
Some of the early results are something like a 6x faster turnaround and throughput from our modern data pipelines. This is a complex transformation that will take us a couple of years to get through the full gamut of it, but we are very excited by the opportunity there. Our commercial engine has been really focused on deepening our client relationships. We had good success throughout the year in extending client contracts without compromising on price. We also managed to increase our minimums that gave us more security in a world that is dynamic. Lastly, we've really reinvigorated the entire company in selling the full portfolio of our solutions that we offer. I started with saying how clients get more value from us the more they use our solutions.
That is absolutely true, but we need to instigate it and expose that to our clients. We have done a great job this last year of starting to do that, and you will see Goran share some examples of the success from that. Winning in an AI world and realizing this generational opportunity I just talked about requires a new playbook. This is what is driving our drive to increase the number of enterprise platform agreements with our clients that give them flexibility and us flexibility. It gives them some confidence in their pricing, and it gives them the ability to mix and match solutions. We are there to lean into that, as I said, because of our open architecture philosophy. Similarly, there is consumption pricing. We offer a whole infrastructure. We have tuned our billing infrastructure, and we have a contracting framework to allow for consumption pricing.
This is not new to us. We have done consumption pricing and analytics for two decades, and that has given us a lot of experience in how to negotiate these contracts and what are the parameters of building trust in a long relationship with client that we are bringing to bear in these new contracts in AI. Lastly, those expert consultants I talked about, those implementation specialists, those analytic specialists and managed services teams, we have taken a portion of them, and we have now trained them up in AI. We have trained them to go into clients and help them build prompts, help them build new agentic solutions. That team, we now call them forward deployed experts, just to pick up the buzzword in the market, and the forward deployed experts are having real success in changing the sales motion when it comes to distributing our AI solutions.
It is early days again, but we are really excited about this because we have the expertise, we have the teams. They are located in the places where our clients are. Our clients already know them and trust them. This is an incredible asset of ours, and the success from our early experience has been very, very promising. The combination of those three engines is really starting to kick in. We now have this operating model where I believe every dollar that we are investing is just being put to work at higher productivity. We are seeing renewed operating leverage. We have a team, a leadership team, and the whole of FactSet is geared towards getting more out of what we have got, and you are seeing that every day in the results.
I am really excited about our ability to capture this generational opportunity with the quality of execution that we are bringing to bear. The strategy I have shared with you, I believe, positions us well to deliver approximately 6% organic revenue growth over the next three years, and also continue to improve margin to get to a level over 36% operating margin in 2029. We are committed to maintaining a high conversion of free cash flow, and together, we think this all brings together a roughly 10% earnings per share growth rate per year for the next three years. That is our baseline. In that baseline, we have captured what we see today. We are really excited about the generational opportunity and the strategy we are pursuing, and we have high confidence that this will lead to client traction, growing client traction, and more. That is not in this baseline.
As we scale the business and as we see clients get more confidence and accelerate their use of AI and adopt more of our solutions, we definitely think there is further upside on top of what we've modeled. Around this time last year is when I started at FactSet, and two things attracted me: an industry at an inflection point and a company with deep embedded client relationships, incredible domain expertise, and outstanding client service. I had seen all of this as a client of FactSet, but now I've spent a year in here and seen it from the inside, and I'm even more convinced. The inflection point for the industry triggered by AI, I believe, is now translated into a generational opportunity for FactSet. I'm excited about this new strategy.
I know all FactSetters are working together to deliver this strategy, and we are keen to show you the details of the product roadmap and how we are bringing this to bear. We believe this strategy has a real chance of creating real client and shareholder value. With that, I'm going to invite Kate Stepp, our Chief AI Officer. She's going to roll a video, and then she's going to be up and talk to us about the product roadmap in more detail and the value proposition. Thank you.
[Presentation]
[Presentation]
[Presentation]
Welcome to the stage, FactSet's Chief AI Officer, Kate Stepp.
Good morning, everyone. My name is Kate Stepp, FactSet's Chief AI Officer. I have been at FactSet for 18 years now, most recently as Chief Technology Officer, so I am very familiar with the value of our data, analytics, and technology stack. My job now is to take that technology stack and evolve it into high-quality AI products for our clients. What you saw in the video was FactSet Intelligence in action, a signal in the market picked up by an agent, enriched with our data and analytics, meshed with our client proprietary data, and turned into an output or a signal that our clients can act on. What I am going to walk you through today is how AI is powering our product engine. I will hit on three key items. First, foundational capabilities. We have decades of FactSet differentiation that is now evolved for agents. Second, trusted AI.
Financial institutions will scale AI that they can trust, afford, and control, and that is exactly what FactSet is building to deliver. Third, a client-focused roadmap. Our AI stack has enabled us to build deeper partnerships with our clients, and what we learn from them informs what we build next. Let us start with the foundation. As Sanoke said, although he loves this screen, it is more than just a screen. Beneath this are a rich collection of tools, engines, and signals, from analytics engines and risk model calculations to transactional platforms. Across our entire platform, we have models and reports that encapsulate rich business logic. That goes across adjustment rules, smoothing algorithms, and custom formulas. Underpinning all of this are the controls that our clients require across security, entitlements, and auditability.
Until today, sorry, until recently, this has really been designed as a human platform for a human sitting at a desk. Today, we are evolving these core assets so that agents can use them too, and that turns FactSet from an interface for humans to a dynamic platform and ecosystem where agents and humans are interacting together. If you look at the middle of this screenshot, you are seeing an agent's chain of thought. Basically, what it takes to complete a workflow end-to-end, you can see our tools and engines being called by the agent. You see calls to our FactSet content as well as client content as well, and that all comes together in a workflow.
On the right, you see an example of our Earnings Intelligence agent, which is a multi-agent system where agents are working behind the scenes in parallel, synthesizing clients' theses because of the fact that the internal research notes are on our platform. Then they are raising those for human review where needed in the context that matters to the user, their coverage list or their portfolio that is also loaded onto FactSet's system. This is just one example of human and agent interaction that you will see today, more examples through the demos. Our core assets, as Sanoke mentioned, have been built over decades of client usage, client feedback, and FactSet iteration on those core components. At the data layer, our large corpus of documents now delivers more insights, from emerging themes to sentiment shifts to market insights that we deliver to our AI applications.
At the analytics layer, the calculators, the optimizers, and the immutable performance history of all of our clients' portfolios, that becomes what makes our analytics layer repeatable and auditable. These calculations matter that they are correct. At the workflow layer, our entitlements and permissioning make secure governed AI possible. It is why one analyst can read a broker or research report at their desk and the person next to them cannot, and those entitlements are also designed to be respected by our agents as well. Our alerting engines bring efficiency to this concept you may hear in the market around always-on agents. Instead of leaving your agents running all the time, our events and triggers now allow them to become motion-activated, so they act at just the right time efficiently. These core assets are not new, but they do deliver new value in an AI era.
There are, in fact, some components that we did have to build new to take advantage of the capabilities that AI now provides. Sanoke started to touch on this earlier, so I will just quickly recap some of the new components that we have built. Across our data layer, we have brought together enrichment, vectorization, and ontologies to deliver more insights, combining not only FactSet data, but our client data as well in a proprietary secured environment. At the analytics layer, we have added coding assistance and hybrid infrastructure to be able to allow our most sophisticated clients to bring together those capabilities, bring the full stack of AI onto their platform to work on top of FactSet data and their client data.
And in the workflow layer, we now have a purpose-built agent platform that brings together all of these capabilities Sanoke was touching on to be able to provide multi-agent ecosystems in a way that's secure and standardized and governed. It's important to note that this AI stack is componentized and modular. It allows our clients to take a workflow solution end-to-end or take pieces of the stack to support wherever they are in the build-to-buy spectrum. On to trusted AI. To set the stage on trusted AI, it's important to acknowledge the pace of frontier model releases. Just to hammer home this point, since the slide was created, more model releases have come out.
If we just took the top three model providers and looked at those, since they have broad adoption across our industry, as Sanoke said, you have anywhere from 24- 30 model releases across this year, shipping every 10 days. If you were to expand that out to look at all frontier model releases globally, you'll get to the hundreds, and maybe at a rate of a daily change or release. Each time one of these arrives, it has to be re-evaluated for quality, retuned, re-optimized, because each one of these models operates a little bit differently, and this can be a choppy experience for our customers. FactSet aims to take this maintenance and verification burden off of our client's plate and deliver a smoother, higher-quality experience by being able to choose and swap in the right models at the right time.
We have dedicated teams who are really thinking about, how do we keep quality rising and costs falling? The way we do this is by evaluating actual outputs and real financial workflows for our clients. This is just a simplified diagram of an agent harness to be able to talk about where FactSet can differentiate. The context window is where everything gets passed to the LLM. Think about the prompt, the grounding data, the related context, the history of a conversation, and all of the results from tool calls. All of that context translates to tokens, and that translates to cost. FactSet adds custom logic based on what we know about the user, what intent they're trying to drive, and what the right model is for the complexity of the job.
The right model might be a frontier model, but the right model also might be a fine-tuned model that's more specific to that individualized task for the user. We see many clients today who are building their own solutions or maybe using frontier solutions, and in order to ground that solution and make the quality higher, they use FactSet's MCP product. That takes the LLM and improves the quality with FactSet's data. Beyond that, though, when we compare FactSet AI and our agent harness against a frontier harness using FactSet's data MCP, we believe FactSet delivers AI that's 10% further lift in that quality, and that comes from our additional content, our intent detection, and having the full context of the Workstation at hand. On cost, we cut a client's inference cost by approximately 30% versus using a frontier model harness while holding the quality steady.
This is an important point because it's easy to use a cheaper model, but you need to keep quality at the level that our clients trust and expect. We do this by optimizing the context for the most important parts before we pass it to the LLM. We can also reduce the number of loops and reasoning costs at each step and choose the right size model for the job. This allows us, again, to deliver the quality that our clients expect at a cost that they can afford. I mentioned optimizing the context. What is the context? Let's talk a little bit about our context layer.
The context layer is the context sources on the left, the content infrastructure in the middle where we're deriving and building on top of that context layer, and then the context activation through our APIs, MCPs, and also agents through our agent harnesses. Our context sources are broad. You have the data that you know and love, fundamentals, estimates, our documents, transcripts, filings, client proprietary, internal research, portfolios, and then also the runtime signals that come through. That breadth is what allows us to answer the questions that clients care about. Can I build a thesis on this? Can I put this in front of a client or an auditor? We believe quality is more than just a score. It's really more about the outcome that it drives and the ROI that we're able to provide.
This derived context layer in the middle is what we curate with industry experts and continuously maintain for expanding AI consumption. Let's go into that a little bit more. Our data mesh is what maps where everything is and how it's connected. Memory lets us personalize that answer based on what's happened before. Our ontologies define what data means, but in the way that our clients actually think about the markets. Because we maintain that and curate that, we avoid the fresh cold start problem. It's always maintained, so we're starting informed. Think of it this way. An ontology is a map of meaning. It synthesizes the entire market, the entities in the market, their relationships, the themes, terms, and concepts that really matter. Our sector ontology was built by industry experts.
For our Earnings Intelligence agents, which you'll see later, across the S&P 1500, we took 81,000 documents from the last five years, mapped nearly 4,000 relationships between companies, segments, and assets, and distilled them into 18,000 market themes. Our StreetAccount analysts and journalists maintain that every day. We believe this is a depth of curated context that's difficult for others to replicate. Let me make that a little bit more concrete. Take Occidental Petroleum. Same FactSet data, same Earnings Intelligence agent. Without the ontology, the agent concludes Oxy is a leverage oil company with mixed results. With FactSet's ontology, the agent immediately sees a Permian Basin pure play, reasons in per barrel costs and well economics. It connects its M&A activity to leverage trajectory and traces the gas supply chain through to utilities.
That is not just a better answer, that is a better category of answer. Here is why that matters, and here is what I want you to remember. Think about a busy earnings season. In the first hour before an analyst opens a single filing, they reconstruct the thesis. They remind themselves what the key debates are, which metrics matter, what the peer set looks like, and what happened last quarter. That is the pre-read. Our sector ontology delivers that pre-read automatically in a single structured call. [Part 35] filings, 10-Ks, 10-Qs, 8-Ks, two years of StreetAccount news, eight quarters of earnings calls, conference presentations, research, and peer companies. Pre-processed, pre-structured, pre-dated. The defining thesis, the live debates across M&A, regulatory and strategy arcs, the dated catalysts, the rounded metrics, the GAAP and non-GAAP pairs already mapped. Ready in roughly 7,500 tokens, or 20,000 if you are including reasoning.
Without that, an agent is reconstructing the same context from scratch, and it reads every source cold in every query. That is approximately 1 million tokens to reach equivalent coverage. All in, this is roughly 50x more tokens without the curation, without the maintenance, and without the industry expertise and judgment that our experts bring to it. Importantly for our clients, this is the difference between a seasoned analyst that has covered a name for two years and one who has never heard of the company. Our proprietary ontologies are just one of the powerful examples of how FactSet is leveraging our broad context, our industry expertise, and our technology across the stack to deliver that higher quality more efficiently. I spoke about how we deliver quality and cost in practice, and Bob is going to talk about control in the next section.
For now, let us move on to what we are building with our clients and the exciting solutions that are in our roadmap. We have talked through how our AI stack helps us build trust, but as Sanoke said, it also helps us build our products. The solutions that Sanoke introduced here, Goran will also speak to how we are selling them to our clients. I want to talk about how they are enabling more deeper, innovative partnerships with our clients. The industry is really changing too fast for our clients to build a whole stack from scratch. They would rather focus on their business and partner with FactSet on the things that we can provide. Here are three examples, each solving a different problem. A hedge fund knew it needed to catch up on AI, but it also knew that its data stack was not AI-ready.
Instead of hiring an entire technology team, it chose FactSet for data management, and with our concordance and our intelligent document service, we were able to take their complex queries from days down to seconds. A global asset manager came to us to speed up custom risk modeling. We were able to componentize some of those different products and bring them together end-to-end, from intelligent document service being able to detect and deliver new emerging themes, which then feed into the FactSet Programmatic Environment, allowing them to define custom risk scores, and then all of that feeding into our risk modeling. This allowed them to take weeks of manual analysis down to hours. Last, a top U.S. bank with a strong technology team came to us looking to how to accelerate a pilot.
With our advisor solutions, FactSet AI, and again, that intelligent document service, we showed how we could accelerate the pilot. We did this by driving adoption in the platform that the end users were already using and bringing in the client-specific context. In each of these cases, a client came to us with a problem, and our rich AI stack was allow us to be part of the conversation and deepen the innovative conversation. The first two here show examples where we were either using our foundation or putting together pieces of a stack. The last one shows how we have the opportunity to really influence the end-to-end workflow solution for the user. That's also the way that we're thinking, and Sanoke started to illustrate how we're evolving the end-user experience.
Just some examples of that that are recently delivered, some visuals to show of what's been recently delivered in the platform. Our wealth agents, that's part of the Advisor Intelligence suite of solutions, has recently been live. Our Earnings Intelligence agent, which you've heard me mention a couple of times, part of Research Intelligence, has also been released to clients. Then coming soon is the Deal Manager, which you'll get a sneak preview later as well, as part of our Capital Markets Intelligence. I encourage all of you to stick around to see the demos later today to get a little deeper into those. Now let me bring it all together. AI is already changing how FactSet builds and delivers, and it's built for how finance works, trusted, purpose-built, and directed by people. Our foundational capabilities give us a head start.
Trust is how our clients adopt AI, and our clients are informing what we build next. With that, I'll hand it to Bob Stolte, our Chief Technology Officer, to go into the operating engine.
Please welcome to the stage FactSet's Chief Technology Officer, Bob Stolte.
Thank you, Kate. Good morning, everyone. I am Bob Stolte, and I joined FactSet earlier this year as the Chief Technology Officer. Prior to that, I spent my career on the sell side, largely at JP Morgan and Citigroup, running large technology organizations. I find it really valuable moving from the client side into FactSet and bringing that experience with us, and I think it's informing us on what level of performance we have to reach in order to meet our clients' needs. Today, I want to take you inside the operating engine and share a little bit of that journey we're on to raise that level of performance. I can organize it into these three themes. The first is simplifying to scale. This is about rebuilding the technology foundation.
I guess the easiest way to say it is it's like doing everything once, doing it well, and then scaling that across the enterprise. The second thing is focusing on control at the core. Our clients trust us with their most sensitive data, and we have to treat safety, security, sovereignty, and resilience as paramount. The third thing we're going to talk about is rewiring and optimizing what we're doing internally with AI, how we build products, how we operate the plant, and how we support our clients to get all of these can drive better outcomes and do that more efficiently. They build on each other. I think we've all seen that if you just take a bad process, you put AI on top, you just get bad outcomes more quickly.
But if you take the time to lay that foundation right, keep controls at the center, then you can maximize the value you get from AI on top. Let's dig in. Simplifying to scale. To me, this is like the foundation for creating operating leverage. Like many companies, FactSet's technology organization— technology platforms, over time, over decades, has become complicated, right? We have hundreds of solutions, tens of millions of lines of code, and some of our processes and tools and solutions over time can be duplicative. We have the opportunity to reduce that complexity. We've put a team together that is realizing that opportunity. It starts with how we build things. This is our development pipeline, the factory that we put software through. We have more than 12 individually managed development pipelines historically.
That's a lack of standardization that compounds the maintenance burden, and it makes it harder to comply with code quality, vulnerability scanning, all of those things, duplicating all that effort. We're following a platform as a service strategy. We're consolidating all of those pipelines and then automating that central capability, providing that back to the organization so that everyone can benefit from that one implementation. Ultimately, all of this is about removing burden from product-facing engineering teams and product teams so that they can focus on differentiated value and not worry about foundational things. Next, it's what we build. We have common capabilities embedded in all of our client solutions that were historically duplicated at times by each team. These aren't differentiating activities. These are foundational things. Think like workflow engines or reporting frameworks, things like that, right?
By pulling those together and collapsing that infrastructure, doing it really well, and then providing it back, we get a higher- quality service, and again, you are taking that burden away from the product-focused teams, the client-focused teams, and then they can spend their energy improving the product and innovating. How we deploy and run. Historically, we have had thousands of cloud accounts, and we have had individual engineering teams all managing their own infrastructure. That is expensive. What are we trying to do? It is complicated. We are implementing an automated containerized deployment and management platform. It reduces the management burden, but it also gives us the opportunity to improve resiliency and to optimize our infrastructure costs by creating portable workloads that we can move across our internal estate and our multiple cloud providers to optimize our expense. To summarize, what do we get by simplifying?
Well, we go faster, we free people up to focus on innovation, and we deliver productivity. Let us talk about control. Our clients do not just use our data. They trust us with theirs. Jake Lilie, CTO for Baron Capital, said it really well. "They are entrusting us with their IP, with their holdings, their research, their risk methodologies." And that level of trust in a partner, in FactSet, that is unique. We have to earn that trust. We earn it through compliance, participating in all of these certifications, SOC 2, ISO. We do it through assessments, where we do vulnerability scans, penetration testing. We do it through participating in industry frameworks. AITEC, SIG, FISD, and of course, we have our internal policies and frameworks that we follow. Examples, application security, cryptography, data governance, all of these.
We bring all of that together, and we demonstrate to our clients in over 2,000 client engagements every year what our control capabilities are. I am very proud of that. But we cannot sit still. The world is changing all the time. Control is critical to scale AI. What does that mean to deliver, right? Well, reliable has to run, right? Has to run at scale. Sanoke talked about some of these things. We are operating at a big scale today, 55 billion Level 1 ticks as our peak last week, 1.7 billion analytic requests, which is trillions of calculations. Well, and we do that reliably. We actually provide to our clients through our platform over 150 measurements of performance, availability, and I picked out one which is four nines for our core platform resiliency over the last six months. So, how is AI changing all of this?
Well, what do you have with AI? You have more change, right? Faster change, and you are going to have more volume. So, we cannot sit still. We have to add, we have to raise our practices around change management, around testing, and around observability and performance in our platform, and we are doing those investments now. Next is about governance, governed AI. Kate talked about quality and cost, and that third element is control, and that starts with the governance for our models and our tools. Policy. On top of that policy, you have that agent platform that Kate talked about. That is where we create the agents and skills. It is also where we curate them. We register everything that is operating in the environment, right? Agents, skills, models, tools, all of it. All of that gets done in the agent platform.
At a physical implementation level, you then have our LLM gateways, our MCP gateways. They give us protection today, and they give us the ability to control those flows. What we're developing on top of that now is this secure agentic runtime environment, where you can, on a policy basis, define what an agent is allowed to do and prevent it from doing things it's not, and on a detective control basis, you can describe and teach what is anomalous behavior and what is not anomalous behavior for an agent that goes outside that framework, you can stop it. We're developing those capabilities now. Last is about sovereignty, right? What do we really mean by this? What we mean is we want clients to have trust that they own their data, right? They have control of it. We provide that today and historically through access control, through entitlements.
We've done that in our multi-tenant core, and we're increasingly giving people those capabilities to control their own data in single- tenant client data meshes. Say that five times fast. We're bringing our capabilities and our data and our control and allowing clients to maintain control of their data in the data mesh. Last thing I'll talk about here is our cyber defense. I think every day we see more and more things happening in the world where, look, the threats are going up. More vulnerabilities are discovered more quickly. The time to exploit is going down, right? Everyone, including FactSet, has to raise the bar on detection and response. You can't bring a human to an AI fight, right?
This is where the threat is machine speed, and so we have to move our detection and our response also to machine speed, and that's what we're investing in right now. In summary, if you take AI quality, AI cost, you add controls, you get trusted AI. We're eating our own dog food. How do we use AI internally? What you see on the screen here is our Spec Studio. This is an ecosystem where our product managers, our designers, our engineers work with fleets of agents to collaborate to significantly reduce the time it takes to introduce new products. The example that's on the screen is from our private capital solution, the AI Doc Ingest capability.
They were adding a feature around allowing people to email documents into the AI Doc Ingest capability, and using Spec Studio, they were able to do that in two weeks instead of like four months or so that it would have taken. That's impressive, right? 75%. You do the math. Another example that I'll give is in our Data Navigator. They released the beta for pricing and the reference data catalog in a month versus a quarter. We're also using AI to power how we run the plan, how we run the data plan. To publish our M&A data product used to take an hour and a half to two hours, and that cut that back to 15- 20 minutes with AI automation of the development process.
We're also actively reducing touch time for our largest dataset, which is our fundamentals dataset, where we're working with the collections team right now to bring the human touch times down in there. Last thing I'll mention is agentic product support. Kate and Sanoke mentioned this one as well. This is where we've gone up this year now to 30% of all of these human requests coming in for help that are getting answered solely by the agent. This is a measure of 100%, no human touch involved in answering these questions. I think this is really interesting. I think we expect that number will continue to grow, and we also expect that the cost of doing it will continue to shrink. This is an example where we're using open-weight models and internal tuning on our datasets to allow us to do this. These are great examples.
Let's talk about what the data actually shows us overall. This graph on the left, you can see our token use has gone way up, 17 x this year. It's probably about right. We're probably not the only firm out there who's using more tokens, right? What I like about this is that you also see that we were able to bend the curve on token spend. Starting halfway through the year, that curve really flattened out as growth continued. You can see we have a 43% reduction in cost per token. So we've grown, and we've grown responsibly. What are we doing with all those tokens? Well, we're up to 60% now of code in the first couple of months this year, the first two months of the year, that are being delivered agentically. What do I think about that number, right?
Sanoke mentioned, I feel pretty much the same way. It's good, right? But it also indicates that there's a lot of opportunity left in front of us to keep going. I had the opportunity to really drill in to the top of that curve, if you will, the largest token users in the firm, and talk to them about what they're really doing and what we're seeing. First thing is, they use a lot more tokens. That's self-evident, right? 6x more than the average. What is less self-evident is some of the value that they're getting, right? We've talked already about time to market, productivity. What we haven't necessarily talked about is things like feasibility. This cohort is able to tackle challenges they literally didn't think were possible a few months ago.
They're also able to do things like drive down runtime costs, being able to spend time, because of the agent fleets that they have, on optimizing code and reducing those costs. Another one that I actually, this one really wasn't intuitive to me, but I thought was really fascinating, is it actually reduces friction and latency in the process. So oftentimes, to deliver a solution, you're dependent on teams or software outside of your immediate control, right? In the old world, what would you do? You'd ask that team to implement that feature for you to make that change. In an AI world, you can actually come up to speed on that code base quickly, create a candidate for release, and submit that back to that team so you're not waiting on them to do that work. You can do that work yourself and really speed up that whole process.
I find that very, very interesting. When you add all of this up, what it demonstrates is we are getting value now, but there is a lot more value to capture as we scale the AI responsibly and we bring more of the FactSetters up that curve, like the top 10% cohort, and get the value. Hopefully this was helpful. Later on today, Josh is going to talk more about how the work we are doing in the operating engine contributes to margin. I am just going to wrap it up and say, as we simplify and scale that foundation, we focus on control with AI, we focus on control, and then on top of that, we optimize with AI, we have delivered value. There is a lot more opportunity in front of us to be better, faster, cheaper, and safer. Thank you very much.
With that, I am going to hand off to Goran, who will talk to you about how our clients are putting all of this to work.
Please welcome to the stage, FactSet's Chief Revenue Officer, Goran Skoko.
Good morning. It is great to see so many familiar faces here today. My name is Goran Skoko. I am FactSet's Chief Revenue Officer. I am in my 23rd year in FactSet, and it has been two years since we last met for Investor Day. At times, it seems like more has happened in these past two years than in the previous 20. AI is changing how clients work, it is changing how they buy, and it is changing what they expect from us. Through all that change, our client relationships remain our core strength. These are durable relationships built over decades, with FactSet deeply embedded in our clients' workflows. It is those relationships and our reach across financial industry that give us a foundation for what is next.
Now, as part of my role, I spend a lot of time with our clients, and what I am hearing is remarkably consistent. As AI moves into their workflows, they need data they can trust. They need fewer partners, not more, and they need or they want FactSet wherever their work happens. It is for that reason that I am more excited than ever for the opportunity ahead of us. As our clients put more data and more AI solutions to work for them, we believe they will need more FactSet. We believe that there is a significant opportunity for us to deepen these relationships, to expand into more workflows, and to serve more users across their firms. Today, I will take you inside the commercial engine Sanoke described and the strategy we have put together to capture that opportunity. I will cover three areas.
First, durable and embedded client relationships that compound over time. Second is strong traction with our new AI offerings. Third, a commercial model that is evolving to capture the value AI creates. Our client franchise is very strong. It is built on nearly 50 years of partnerships and trust. As Sanoke highlighted, it is also very diverse. We have over 9,000 clients in over 80 countries, and that gives us a broad foundation across markets and geographies. Many of our clients have been with us for years, and 75 out of our top 100 clients have been with us for more than 20 years. That durability reflects the trust that we have earned, that continue to earn, and you see it in our ASV retention. It is greater than 95% overall, but it is over 99% for our top 100 clients.
Our embeddedness expands beyond data that is deeply integrated, as Sanoke described, into extensive product lines that our clients deploy to resolve their problems. Now, let us take a look at how client obsessions and our FactSet service model fuels our land- and- expand strategy. Expanding existing client relationship is by far the biggest growth driver of our ASV growth. I have to say growth a lot of times, because as you can tell, it is part of my expertise. Every relationship has a starting point. Over the past six years, we have acquired more than 3,000 clients, either organically or via acquisitions. As we onboard these clients, we focus on adoption.
Our customer support, our client service organization, our consultants work very closely with our clients to make certain our tools are implemented properly, they are solving problems we promised our clients to solve, and in that process, we earn trust. What frequently starts as a one-solution relationship then expands into additional teams, workflows, and geographies. As you can see, we grow with clients at every stage. What each bar in this graph shows is how much net ASV clients had when they expand with us. We used the one product line as the baseline, and that is 1x. On your left, you see when the clients that use two to four product lines, they expand by 1.5x that baseline. As that embeddedness grows, you can see how the growth compounds.
Clients using five to seven product lines, when they grow, they grow 2.5x baseline, and deeply embedded clients with 10+ product lines grow 35x. That is the power of our model. As clients adopt more of our solutions and see greater value from FactSet, our relationship deepens, and we are creating more and larger opportunities for growth. Let's see how these relationships evolve in practice. We will take a look at three relationships, one each in North America, EMEA, and APAC. These are three very different journeys, but they all have one common thread. They are longstanding, long-lasting relationships that continue to deepen and create expanding value. First is a North American client. They have been with us for 20 years, and their ASV has grown at 14% a year over the past decade. This relationship started in banking and research and then expanded to wealth management.
We quickly became their enterprise wealth platform, further expanding across geographies, departments, and workflows. Second is a leading European asset manager. This relationship started as a one-point solution 20 years ago. It is now a multifaceted relationship spanning their asset and wealth management divisions. Last year, our ASV with them grew 22% as they selected us as their managed service provider for their performance function. Lastly, a leading APAC asset owner, 16-year relationship that grew 15% last year as they chose us as a partner for their AI-ready data. These examples show how enduring relationships can become engines of growth, even for some of our largest clients. What is next? How will our AI solutions shape the next stage of our client relationships?
Kate and Bob walked you through the transformation of our product and of our AI product and technology, the architecture, our vision, and ambition. This slide shows you how our sales teams are already bringing all of these tools to market. We have significant traction across all three layers of the stack, from connected data to advanced analytics, and all the way to intelligent workflows. Let me walk you through how clients are already engaging with us at each layer of this stack. At the data layer, early strong contribution came from MCP, and that momentum continues. We launched our MCP less than a year ago, and since February, we have seen significant growth in adoption of our clients and even faster adoption across active users. Let me highlight two important data points.
One is investor often asks us, whether clients will use the Workstation less as MCP adoption increases. What we are seeing is the opposite. As FactSet content becomes available across more user workflows, they are turning to Workstations more often for deeper analysis. What we are seeing is that the clients or the users that have adapted MCP have grown their active engagement with FactSet Workstation from 89% to 92% post that adoption. MCP is also having a positive impact on the ASV growth for clients that adapt MCP. In 2026, clients that bought MCP solutions from us grew 10% overall, compared to just under 7% for those that do not have MCP. We monetize MCP in two ways. First, MCP often drives broader enterprise agreements with additional revenue as usage grows.
And second, we offer MCP through a subscription that includes a base level of access to content, and then consumption-based pricing kicks in after that. So at the advanced analytics layer, we also see very good traction there. A leading wealth management solution wanted to engage deeper with their end clients, wanted to deliver industry-leading performance analysis and insight, and with the ultimate goal of driving or improving new business acquisition rates. So they turned to FactSet as a trusted partner. We engaged on the journey with them with a small number of portfolios. We scaled it significantly since, and we continue to scale it, with the ultimate goal of serving over 1 million accounts for this particular client. So by scaling our proven institutional-grade analytics and reporting capabilities, we broaden how we serve wealth management clients. We have grown significantly in this space.
We have over 130,000 advisors using FactSet solutions every day. So that gives us a great foundation to expand into additional workflows for them, and portfolio workflows are a perfect area for us to expand into. In summary, we are bringing institutional-grade analytics to more advisors and their clients, helping them grow their business, but at the same time, we are generating new growth opportunities for FactSet. And lastly, we are very excited about our recent success at the intelligent workflow layer. FactSet AI for Banking is deeply integrated into the way our clients work. It brings together all layers of our AI stack, data analytics, and agentic workflows. In a secure single-tenant environment, bankers can combine their firm's proprietary data with FactSet data, meshing it together so they can use it in their workflows.
We are currently in the process of deploying FactSet AI for Banking to 3,000 bankers at a major institution, and the user feedback is very positive. Tasks that took bankers two hours to complete now take minutes. And importantly, the output quality is higher than what they could have done on their own. That is what they tell us. FactSet AI isn't just accelerating the work, it is elevating it. This sale led to 28% expansion in ASV with this client, as Sanoke pointed out, but the numbers only tell the part of a story. A very important data point is that this client did not have to adapt a brand-new AI platform. We simply flipped a switch and enabled our AI solution on FactSet platform, the product that their bankers use every day. FactSet AI is now seamlessly integrated into their workflows.
My point is that our extensive presence in banking creates a significant distribution channel for us and a significant upsell opportunity for our workflow solution. And overall, this deployment is a great example to demonstrate how FactSet can solve real client problems at scale by combining all layers of our stack. Now to turn everything we have spoken about, our strong client relationships, connected data, and AI into growth, we need the right commercial and go-to-market model, and that is exactly what we have been building. First part of the model is longer agreements and longer and broader enterprise agreements with our clients, and we are seeing very good traction there. I believe the reason is because our clients' interest and our interests are aligned.
This gives them broader access to FactSet platform across their workflows, better visibility into their future costs, and at the same time, it gives us a more stable revenue base that continues to support our growth. At the same time, we are evolving how we monetize the value we deliver with consumption pricing. Consumption pricing aligns what clients pay with their use of FactSet's product. It gives clients flexibility to interact with our products on demand, and it gives us the opportunity to grow as that adoption expands and their usage grows. Sanoke pointed out that this is not a new motion for us. We have more than 80 of our top 100 clients utilizing or having consumption-linked arrangements as part of the relationships, mostly relating to our portfolio products as well as to some of our data products.
That experience has given us experience in how we negotiate these type of agreements, how we track and measure usage, and very importantly, it has developed trust with our clients in terms of how we manage these type of arrangements. Lastly, we are also evolving how we go to market with our forward deployed experts. We are embedding our client-facing talent into clients' organization so that they can solve problems, they can build solutions, and they can create value and not just support and cross-sell our products. We have a very strong foundation in place already here. We have hundreds of consultants that work in our client offices every single day. Recently, we have been upskilling a subset of them in each region, and their engagements with our clients are already producing some very promising results.
We're seeing faster and better close rates when our forward deployed experts engage with the clients than when they are not. How does this show up in practice? First, we recently transitioned a top five client to a five-year enterprise relationship. This gave them a broader access to our solutions, our content, and they're using it across their entire banking workflow and across their AI solutions. There's a commitment to 17% contract value ramp over five years, and a potential for further upside with our new AI offerings. The second example is asset manager with frequently changing and expanding data needs. We added a consumption pricing-based component to that arrangement. They gave them more flexibility to access our APIs, data feeds, and MCP, enabling both their quant teams and their AI development to proceed faster. The relationship grew by 9%, and there is growth in the upside with consumption.
There is a base set of content included in the agreement, but if they grow beyond that, we will realize further upside. Last is a leading asset manager, a very large organization, turned to us to help with the deployment of their internal AI solutions. They wanted to improve the results that that solution was delivering. Our forward deployed expert worked very closely with them, connected our MCP with their AI solutions. They achieved the results they desired. This led to a $300,000 sale of MCP for us, but more importantly, it helped grow that relationship by $1 million at renewal. Across all of these three examples, you can see our commercial strategy taking shape. We adapt how we deliver and price our solutions based on clients' individual needs, and we create new opportunities to grow these relationships.
Let me conclude by saying our trusted client relationships and early momentum of our AI solutions are creating, give us a real opportunity to grow with our clients. Our evolving commercial model is designed to capture the value we deliver as clients put FactSet to work across more workflows. As I frequently tell our clients, don't take my word for it. Let's hear what our clients have to say about their relationship with FactSet. Thank you.
Timely, accurate data helps us make the most important decisions in our clients' lives, and FactSet is a great partner who provides that timely data on a regular basis and provides a level of service that's very valuable.
FactSet is the beginning, middle, and end of our day.
It's mission-critical because it sits at the heart of everything that we do.
We use FactSet right across the investment life cycle.
What started initially as a replacement to an existing provider, has become a long-term partnership that we value.
FactSet touches multiple parts of our investment process. Equally important though, it is also the backbone of our risk analytics and performance measurement capabilities.
What attracted me to FactSet was the breadth of the services, the breadth of the data, and the recognition of the brand.
There's almost no parts of our business, whether it's in our commercial bank, investment bank, in our wealth management business, FactSet's been an incredible partner and has helped us serve clients, and we really appreciate and value the relationship.
FactSet's a trusted partner because of the people. The technology is brilliant. The people are even better.
For me, the value of FactSet goes beyond the technology. There is real value in the global expertise and perspective they bring. We need FactSet to understand our operating model, where we are heading, and what we need from them, because that gives us a strong foundation to build from.
When we come to them with a complex issue or a question or trying to work through something, they don't come back trying to sell us a product. It's, "Let's sit down, let's understand what you're trying to accomplish, and then try to see the best way to get there."
Service is key. They have been fantastic with us.
They have really come to us with a white glove service model, where they brought their top leaders on their side from the different business lines.
FactSet Professional Services brought in their A-team to help us with our migration and journey all throughout this.
FactSet offered us an enterprise contract that aligned with what we are trying to move towards, which is enabling FactSet across the entire firm without an additional line item being added each time a new hire is added or we needed a product line.
This whole combination of domain expertise, responsiveness, and the partnership mindset has made FactSet a trusted partner for APG Asset Management.
AI wins or fails based on trust, and FactSet does have incredible data across a spectrum of different sets.
If you can't trust the data, that's basically a recipe for disaster. You're asking an investment person to make an investment decision based on the data.
The LLM is only as good as what data it has access to.
What gives me confidence in FactSet's approach to AI is that the fundamentals haven't changed. FactSet has always been a strong data-driven organization, and we were having conversations with them about AI well before the current wave of interest.
They were one of the first vendors that we saw that brought an MCP to play.
FactSet's innovation in AI is today, tomorrow, next week, as opposed to six months ago, which is an eternity in the AI space.
FactSet's a good fit in this space because they understand that it could be either build or it could be buy. They're providing us with those APIs, those MCP connectors, so we can build it into our own existing tools that we have.
It is also the continual innovation that we are seeing from FactSet, and we are seeing it now with new AI capabilities that are being rolled out. It is just a continuous cycle of innovation.
What really excites us even more is what FactSet continues to innovate, right? Particularly in the areas such as AI.
FactSet's AI Foundry team is very unique. It gives Causeway the opportunity to partner with other like-minded leadership in the AI space, and I do not think there is any other vendor that we are working with that offers that type of level of partnership.
We were really able to partner with FactSet to ultimately build from scratch a new capability, systematic credit research, that is fully cloud- native and all in less than a year.
FactSet makes possible for Causeway the ability to do more with less. It is really the ability to be a force multiplier.
It is really nice to work with people who really have spent the time and really understand our business.
FactSet helps us turn trusted data and technology into new investment capabilities that we can offer to clients, which ultimately is the arena that we play in.
FactSet has been an incredible partner.
Phenomenal.
It's also the backbone.
Incredible data.
A white glove.
Force multiplier.
FactSet's mission critical.
Long-term partner.
Fantastic support.
A trusted partner.
The show will resume shortly.
[Break]
Our program will resume momentarily. Please make your way back to the main stage. Please welcome to the stage, FactSet's Chief Financial Officer, Josh Warren.
Thank you. All right. Okay, come on back. All right. Thank you. Welcome back. It's great to see so many familiar faces. A lot of your heads down in your laptops, but still great to see you. Great to see so many people on, or can't see you, but understand that there are over 125 people who've joined us via the webcast live, so thank you for joining us. It's great to be here at my first FactSet Investor Day. I'm Josh Warren, CFO. We're excited to reintroduce you to FactSet today. This morning, Sanoke walked through our long-term strategy, our vision for becoming the most trusted financial intelligence platform in the industry. You've seen exciting detail about our product engine, our operating engine, and our commercial engine. This gives us the confidence that we have the capabilities to capture the opportunity in front of us.
What I would like to do is to try to bring all of that together and talk about how we create value for FactSet shareholders. Maybe let's first set the scene. Today's all about the forward, but let's set the scene. During fiscal 2026, we developed our strategy, we evolved our operating model, and we improved our execution, specifically our financial results. These were at or above all of our metrics. Consistent with being a stable compounder, we grew our free cash flow per share by more than 20% during fiscal 2026 to $19.32. It's this combination of strategy and execution that's intended to drive compelling returns to our shareholders going forward. We expect these solid results to continue, so we're taking the opportunity to reaffirm our outlook for fiscal 2027.
What I'm excited to do is to step back from the cadence of quarterly calls and take a longer-term perspective about why the value we deliver creates value for FactSet shareholders. We've talked about AI extensively this morning. AI is having a positive impact with our clients and in our own operations. As Goran said, AI is strengthening client relationships. During fiscal Q4, clients with our AI solutions grew their ASV with us twice as fast than the rest. AI is contributing to new revenues. AI solutions accounted for more than 10% of our net new ASV in fiscal 2026. I like that phrase Bob used, dogfooding. Maybe sip our own champagne. AI is already helping our profitability. For example, as you heard earlier this morning, our Client Assist agent is lowering our cost to serve, creating favorable incremental margins.
AI is helping our OpEx efficiency by both improving and enhancing our product and our software development. Based on our market position and on the evidence, we believe AI is a tailwind for FactSet. Our strategy is to capture the value of this opportunity. I am going to outline three themes in service of my objective here today to talk about how we create value for FactSet shareholders. Our consistent, durable revenue growth across a variety of market environments and technology transformations, the margin expansion opportunities that we see from a combination of operating leverage and productivity improvements. Then, because growth and efficiency translate into earnings and free cash flow, our disciplined capital allocation framework is designed to ensure that cash will be allocated to create shareholder value. I am looking forward to providing detail on all three, but let me start with durable growth.
As Sanoke and Goran detailed, we have a strong and resilient client foundation. They use our platform, it is purpose-built, and it has continued to evolve alongside our clients over years, becoming deeply embedded, deeply rooted inside of workflows. Because of this reinforcing foundation, FactSet has grown revenue every year, growing it by more than 50x from the IPO 30 years ago to today. A lot has happened in the world during this period. There have been strong markets, there have been challenging markets that have strained FactSet's client base. However, FactSet grew through the dot-com bust, the global financial crisis, and through the COVID-19 pandemic. This is because FactSet is there to support mission-critical business functions for our clients. It is essential to how our clients operate.
It is this resilience and consistency of our business that flows through to our financial results, which have driven strong returns to our shareholders to date, and we believe there are significant opportunities ahead. As Goran detailed, it is because of our foundational strengths built over decades that we have a diverse set of clients across the financial services industry, including many of the largest, most demanding, most highly regulated, systematically important financial institutions. We serve them in multiple parts of their operations, and this is particularly true of our top 100 clients, who comprise some of our longest and most strategic relationships, and approximately 40% of our recurring revenue base. Our recent results are an application of our durable growth model. Q4 fiscal 2026 was a record quarter for FactSet. We grew organic ASV more than any quarter previously.
While we have shared multiple case studies with you this morning about the trends that we see, providing color on what we are seeing actually with clients, we are also bringing on new clients to FactSet. New clients typically start with a small share of wallet, and it is that which creates our runway for expansion. These client relationships, they season over time. Clients integrate FactSet into their workflows, and 75 of our top 100 clients, including nine out of our 10 largest clients, have been with us for more than 20 years. Because our base renews reliably, new ASV adds to growth rather than replacing lost revenue. Then, as these relationships continue to evolve, we are laying the foundation for even more growth. Those top 100 clients, through favorable enterprise platform arrangements that align interests, including new purpose-built products, these represent some of our most exciting growth opportunities.
And it is all of this combination that supports our outlook of approximately 6% organic revenue growth over the medium term, with significant opportunities for upside, which I will detail shortly. You have seen this slide a couple times today, but we have the ability to monetize every layer of the stack. We deliver connected data. We deliver it directly. We deliver it through an open architecture posture with a client-focused and a client-led roadmap. With data as the foundation, we deliver advanced analytics that process that data into insights. Then with those insights, our data and our analytics become nested into workflows that get something done. Embedded and connected data, advanced analytics built on that data, workflows that orchestrate an action based on those analytics, based on that connected data, are the foundation for strong and growing relationships.
But first, we are working to ensure that more and more of that growth translates into earnings. Talent productivity is central to these opportunities. From fiscal 2020 through fiscal 2025, headcount at FactSet grew every year and grew at approximately 4%. FactSet added headcount to support growth. In fiscal 2026, headcount declined 3%, our first decline during the decade. But we did this while delivering record organic ASV and preserving the quality of service that our clients value. Our productivity initiatives are intended to let us grow without adding headcount at the same pace, which provides a pathway for increasing operating leverage. And we expect to continue to invest to drive growth, and at the very same time, our strong foundation and strategic positioning can drive margin improvement beginning this fiscal year and continuing in the years ahead.
We have discussed the P&L impact of performance-based incentive compensation arrangements a couple of times with you. This was certainly on display in fiscal 2026 because we outperform on ASV, and ASV is recognized as revenue over time. Expenses are recognized upfront. There can be a temporary margin drag. However, a large percentage of our cost base is fixed. And that is why we believe these productivity initiatives can drive both growth and margin. And we have many levers at our disposal, many levers for margin expansion, and we are starting to see the strong early returns of adopting AI in our own initiatives. I am going to detail three levers in particular. First, operational streamlining. These include limiting G&A growth, as Bob described, simplifying our platform. In fiscal 2026, we achieved $30 million of gross savings, and we expect these savings to continue in fiscal 2027.
Sales and marketing efficiency, where we are continuing to streamline, to delayer how we are organized. Our client service is more efficient while maintaining the quality the clients expect. We expect continued improvement here. Then third, as Bob described, we are transforming how we work to enhance our profitability with limited incremental cost, and that flows from having a scalable infrastructure. It is because of these three initiatives that we have high confidence in creating more than $100 million in gross savings in the coming years, with line of sight into more. It is this that creates the capacity for both organic investment and margin expansion at the very same time. And we see clear opportunities to drive margin efficiency, margin improvement, to drive efficiency while we invest in our growth.
For you to make that easier to track, we are introducing a more functional view of our cost structure beginning in the first quarter of fiscal 2027. In addition to aligning with how we operate the business, we think this leaves you greater transparency and improved comparability with some of our peers. This new view shows where we are driving leverage and where we are investing. For example, we will continue to invest in tokens, in compute, in new content and new coverage, particularly in private markets. At the same time, data ingestion, common tools, automation opportunities, efficiency across our G&A functions, that will drive operating leverage. We expect to reinvest the capacity we free up in delivery and anticipate returning a portion to the bottom line. Our disciplined capital allocation framework is powered by consistent free cash flow generation.
It is a hallmark of our operating model, and we believe it positions us to deliver for clients and for shareholders. Free cash flow has grown steadily from $428 million in fiscal 2020 to over $700 million in fiscal 2026, a compound annual growth rate of almost 9%. Three characteristics drive this consistency. First, our foundation. Revenues are predictable, flows from longstanding client relationships. FactSet is asset- light. CapEx is typically less than 5% of revenues. We have high free cash flow conversion, typically exceeding 90% of earnings. Along with predictable free cash flow generation, our balance sheet provides capacity and strategic flexibility. Our gross leverage is 1.5x, approximately a full turn below our historical target. We expect to maintain our investment- grade status with Fitch and Moody's both recently reaffirming us with a stable outlook.
I will note that in August, we upsized our revolver capacity from $1 billion to $1.5 billion, and we did that while extending the duration of our credit facilities, enhancing our maturity profile, lowering the rate in the event of a draw. A larger facility, a longer duration, and more favorable pricing, those three typically don't go together in the credit world. However, for us, it reflects the strength of our foundation. It gives us flexibility as we deal with a 2027 maturity that has a very low cost of carry. So, those are the sources. We aim for every use of capital to contribute to shareholder value creation. We aim for our capital to compete on a risk-adjusted return basis, investment first, and then capital return. We evaluate both organic and inorganic investments based on contributed return on invested capital relative to an appropriate hurdle rate.
We look at project IRR. We look at payback period, given our client-focused roadmap. We evaluate capital return based on accretion to EPS and on free cash flow per share. The aim of our framework is for capital to flow to the opportunities with the highest risk-adjusted returns, aligning our interests with our shareholders. We see the greatest return on capital investments in our transformed operating model. You have seen it on display today, our reinforcing engines. These enhance client relationships, they strengthen our core capabilities, they expand our product offering, they position us for new AI opportunities. They position FactSet for long-term sustainable growth. Because of our client foundation, we believe we can drive further compounding over time. Where we do look to expand, we apply a rigorous buy-build partner analysis.
Any acquisition that we pursue, it has to flow from the highest area of strategic fit in our platform. We have a deal pipeline. That deal pipeline typically includes trusted partners already integrated with us, either technically or commercially. We expect to be disciplined with that pipeline, and our September acquisition of the BCC Group provides a good example. In that instance, we worked together on multiple deployments for multiple clients, and we can get more done together. FactSet is going to be disciplined. We are going to be very disciplined as we look for opportunities to create shareholder value where we are an advantaged acquirer. In addition to investing in the franchise, where we focus on growth, we return cash to our shareholders. We expect to do so through both a predictable and growing dividend and through repurchases, further compounding value on a per-share basis.
Since fiscal 2020, FactSet returned over $2.8 billion, approximately 20% of our market cap, to our shareholders. We grew our dividend for 27 consecutive years, and that is because of the strength of our business model. Regarding share repurchases, the market dislocation of fiscal 2026 led to elevated buybacks. In fiscal 2026, we more than doubled our repurchase activity to $644 million from $300 million the fiscal year before, at an average share price of approximately $247. Over the long term, we expect to return to our programmatic pace. However, we continue to be opportunistic. Meanwhile, at the same time, our weighted average shares outstanding have declined by approximately 5%.
It is this combination working together of a strong client foundation, margin opportunities that flow from operating leverage and productivity initiatives, and a clear and prioritized capital allocation framework centered on shareholder value creation that underpin the confidence that we feel in our medium-term outlook. While our focus will be on compounding free cash flow per share, we expect our franchise to deliver continued organic revenue growth, improving profitability, and continued high free cash flow conversion. We expect to deploy that cash generated in line with our capital allocation framework, driving shareholder value and generating approximately 10% annual EPS growth through fiscal 2029. I want to pause here for just a moment. Everything here that I have described so far, that is the baseline. You have heard it today. We believe AI is a generational opportunity. As Sanoke described, the industry that we serve has a high bar for trust.
We believe, based on what we see, AI will ultimately drive value on top of our baseline. As Kate described, quality is paramount in a world of AI. Clients demand the right answer as efficiently as possible, and clients want to remain in control of their IP. We believe that we are well-positioned as the industry moves past what it is today, experimentation, emerging adoption, into broadening adoption, and then ultimately into agents providing real, meaningful work for their clients. We expect to continue to do more with our deepest and longest-standing relationships, and we will keep you updated on our progress with a clear set of indicators that will demonstrate our ability to grow beyond our durable and compounding baseline. To summarize, to wrap it all together, our focus is on compounding free cash flow per share.
FactSet has a long track record of profitable growth, growing with our clients through market cycles. We plan for this to continue, to continue to grow with our clients, and our algorithm is expected to deliver approximately 10% EPS growth per year. On top of that baseline, the potential for any acceleration in AI adoption gives us the potential for upside. We believe that we are well-positioned to capture the value of that upside and deliver long-term value for our shareholders. I appreciate everyone's attention this morning. Thank you for joining us. With that, I will turn it back to Sanoke for some closing remarks, and then let us open it up to Q&A.
Please welcome to the stage FactSet's Chief Executive Officer, Sanoke Viswanathan.
Thank you, Josh. You have heard over the last couple of hours why we believe FactSet represents a unique shareholder value creation opportunity now. We have strong foundational strengths. These strengths include our broad distribution, our connected data, our embedded workflows, and our unique service model. Building off these strengths, we have created a strategy, a strategy that is going to enable us to drive confidence in the adoption of AI in financial services. The strategy has three parts. Part one is to connect our data and mesh it with client data to create unique proprietary context. Part two, to deliver advanced analytics to large numbers of new clients, new functional domains, and use cases. Part three is to transform the current user interface and the user experience of over a quarter of a million installed workstations into a new powerful human agent orchestration platform.
The value proposition that we can deliver on quality, cost, and control, we believe will engender trust in AI. We believe we can create that trust and reinforce the trust that clients have with us over decades, and that proposition is already starting to show. Clients are adopting our products, and when they adopt our products, we see real monetization follow. We have made solid progress, and we are well advanced in our operating model transformation. Our product, operating, and commercial engines are working in lockstep, and we are seeing opportunity for much more. The commercially focused product roadmap of AI products, supported with the internal agentic transformation of the enterprise, and our restructuring of our commercial model to include enterprise platform agreements, consumption pricing, and forward deployed sales motions is well underway. By now, you have seen the depth and the experience of our management team, and the others are here.
I encourage you to meet with them over lunch. We believe we have a team that is focused. We are executing with intent, and we are very focused on driving productivity and renewing operating leverage. Combined with our focus on maintaining high conversion of free cash flows and disciplined capital return, we believe we represent attractive shareholder value in the coming years. We are well on our way to building the most trusted financial intelligence platform. Every FactSetter is excited to be here today, and we are pleased to be building this solution now, and we believe there is opportunity for real shareholder and client value creation. With that, let me invite the rest of the team back up on stage, and we will open the floor to questions. We are going to bring some chairs up. Kevin, will you moderate?
Yes. We covered a lot this morning, so now we want to hear what is on your mind. We have the full team here. We just ask that if you raise your hand, we have a couple of mic runners coming around. Once you have been selected, just state your name and firm, so that the team knows who they are talking to.
All right. Thanks, man.
We will start here with Kelsey over here.
Hi. Good morning. Thanks for the super helpful presentation. This is Kelsey from Autonomous. As you think about positioning FactSet as a one-stop shop across data, analytics, intelligence, workflow solutions, it does seem like there's a heavy emphasis on the services element that you're providing. If we look out the next 5- 10 years, is it still important for FactSet to own the user interface, especially in the agentic AI world, and is it still important for FactSet to invest in proprietary data?
Sure. Two parts to that question. One, increasing services element and the importance of owning the user interface versus proprietary data. I'll start by reiterating, we've always had a strong service element to our business, Kelsey, and that's one of the things I was trying to illustrate with the depth and breadth of the expertise that we bring to bear in our services model. That's always helped us build these long-term relationships and bring the best out of our products. The model that we anticipate going forward is very similar, in that we have these sophisticated AI solutions being rolled out into the market. These require a level of forward deployed engineering and forward deployed expertise, and that's precisely what we are leaning in into. The user interface, we believe, that we have developed, and you'll see shortly over lunch, is tremendous.
It gives the value to clients in the way they require. It benefits from the context and the learnings that we've had over the years of, as I said, over a quarter of a million end users and their feedback over decades. We believe we are good at building and launching excellent user interfaces, and this technology transition with agentic is just the most recent one, and we are taking advantage of it to create really powerful human agent interfaces. Our data will always remain a very strong connected data foundation. We believe we have lots of unique datasets, and the true proprietariness comes when we mesh our data with client data and create a unique knowledge base that is proprietary to the client. That's where the real unlock and the foundational It's a foundational building block of the value that the client is able to realize from it.
Can we go to Toni in the middle?
Thanks so much. Toni Kaplan from Morgan Stanley. I wanted to ask, in the case where customers are consuming more and more of your data in their own AI environments, I guess, does the 95% retention rate, is that still going to be a key part to the FactSet story? Of course, if this is where they can get the data versus others and it's proprietary, then I totally get that the retention's going to be high. But I guess in the sense of, does the new AI world sort of lead to less, "I'm using FactSet," and more just, "I'm using a data source"?
Yeah, it's a great question. I would say customer behaviors are still evolving on this topic. What we hear from customers, and Goran, I'll probably look to you to add some more color on this, is clients have tried using different datasets. I think, obviously, the most fun stories for us to hear are when they've tried other datasets and things have broken, it hasn't worked very well, everybody's up in arms, and they come back to FactSet and said, "Look, we need your data back," right? This has happened multiple times through the year. It's not a surprise to us. There is a tremendous amount of quality engineering that goes into our data products. It's not just about how much data and how many years. It's the quality, it's the nuances, it's the very specialized extra metadata that we add to it.
There's a lot of nuances to data, and I'll maybe ask you to seek out Justin Strand or John Costigan later. They'll go wax eloquent about it. I'll just summarize to say, we believe the ASV retention number really is not going to be affected. In fact, it's going to probably improve in an AI world, because customers really see the value of what our data does to the quality of the AI output.
We'll stay in the middle with Alex.
Thank you. Alex Kramm, UBS. Sorry if I am asking a couple, and these are financial questions. First on the guide, you basically said it is a baseline. Maybe the question is why. You are talking about all this confidence that you are having here today on the AI strategy, so shouldn't that be your base case. Or maybe just talk about, I do not know if it is upside potential or the confidence level. Is there still a risk that you see into the downside. Then, Josh, two very quick ones.
You did not guide ASV, so is that a similar number. Does the EPS guide include share buybacks, and how much. Thanks.
Let me start, and then Josh will add.
Sure.
First, we are excited about the strategy, Alex, and we wanted to share it with everybody. We believe we have a unique advantage here and a unique opportunity to build something very special. That is what the company is focused on. At the same time, as we do guidance, we prefer to guide with high confidence, and we are guiding based on our bottom-up modeling, and that is based on what is already observed. As there is some AI that is already observed, we have high confidence in that, and that is what is reflected in the baseline. We do think there is opportunity for upside. We will keep you posted along those leading indicators that Josh talked about. Every quarter, every year, we are going to keep you updated on the traction we are having.
And maybe just to pick off — we won't call them two and three— 1B and 1C, Alex, of your questions. Share repurchases, yes, included in consistent with our capital allocation framework. You should expect that to continue to be a part of what FactSet does going forward, part of our value creation algorithm. And the other one was ASV and revenue. ASV is NTM revenue. And fundamentally, our expectation is those will continue to stay in tight correlation. What we haven't modeled, to Sanoke's point, is any dramatic acceleration from consumption-oriented pricing, which would cause those two to diverge. If you want to think about ASV as forward revenue, that's a very reasonable way to think.
Let's go over there to Shlomo.
Hi. Good morning. Shlomo Rosenbaum from Stifel. I have a few questions for Sanoke. First, I wanted to ask you, with everything going on in AI and the ability to significantly increase the analytics, can you talk about where you expect the analytics pricing to go? In the consumption-based model, how do you expect that to impact the revenue of the business over time and how that's going to work? Then, just a broader question. In terms of what I've observed in FactSet, let's say over the last 17 years, the pace of change is dramatically different than what I've seen any time before. Can you talk about some of the management challenges over there in terms of addressing all that while the company is trying to grow?
Also, just on the systems, the underlying systems that you need to be changing, and how do you ensure that they're working and you're not having any breakdowns as you're trying to change it while you're going 75 mi/h ?
Sure. Lots of questions in there. I will start with analytics. I do think the unit cost of producing analytics is absolutely coming down. This is one of the benefits of AI. The picture I showed you of FactSet Programmatic Environment, which you can check out for yourselves outside, the quality of our new agentic stress testing that you will see outside in our portfolio analysis tool, these are all things that historically would have taken teams of PhDs months to build, and now you can do it over a weekend. The unit cost of production is coming down. We believe that is going to translate into significantly higher volumes of analytics being actually consumed in the world, because we are not really at the efficient frontier of analytics consumption anywhere, not in wealth management, which was the example Goran shared earlier.
There is a big opportunity for the overall opportunity to grow, but the mix will be a lot more at lower unit cost. So it is like Jevons paradox, right? You are expecting to see an explosion in activity that will more than overcome any kind of price compression. When it comes to the pace of change, look, it has been exciting. I mean, I cannot go fast enough. I think all the FactSetters here certainly will agree with me that it has been a lot of change. But as it has been for the entire industry, our clients are changing fast, our competitors are changing fast, and we are just changing at the pace at which we believe we can execute our strategy thoughtfully while keeping things safe and sound.
We are, as we showed earlier, we are supporting mission-critical processes, and we cannot afford to allow any of those to break on any given day. If we break, that has to be responded to instantly, so we have to have the high observability. We are investing in all of these kind of capabilities, resilience, scalability, and reliability.
Can we go to Surinder in the back.
Surinder Thind with Jefferies. Following up on the AI upside here, how should we think about the timing and what you're seeing at client behavior? You hinted at certain metrics that you've provided about some of the uplift that you're seeing at some of the clients, but how pervasive should we think about that opportunity and the client's readiness to start to rebuild some of these workflows over the next year, the next two years? Because if 10% of growth, I think, is already of ASV growth is already AI, what should that evolution of that look like over the next couple of years?
Yeah, it's a question we ask ourselves, Surinder, and I would say we are going to be client-led here. We can tell today in the market, as we talk to hundreds of clients, that the binding constraint to adoption is not the frontier model capability anymore. It really is internal organizational ability to absorb, and it's trust. So we can help solve the trust equation. Everything this morning we've shared with you is all about that. The internal organizational challenges of moving from individual user adoption and change to what they do one-to-one versus a real transformation of team processes, that's what our clients are working on. And at the pace at which they are able to evolve that is the pace at which these solutions will scale. We will provide the trust equation.
We are comfortable that there is a lot of traction with that, and we'll continue to keep you updated with those metrics. What we are keen to show you is a steady pace of new product introduction, the adoption metrics of that, the engagement metrics of that, and when they convert to monetization.
Go down in the front to David.
Thanks. David Motemaden from Evercore ISI. Just another AI question on the FactSet harness. You had said that that's 10% higher quality than the frontier lab or horizontal harnesses, and you test that every week. I'm just wondering, how has that changed? If you're testing it every week, I'm sure you can track it as new frontier models get rolled out. Is it 20% going to 10%? Is that 10% stable? Also, I guess, do you have a view on how your harness compares to a vertical AI solution harness?
I'll start, and Kate can add to that. First, I'd say absolutely, we see that continued uplift, right? It's not that it used to be a much bigger lift that is narrowing. There are multiple reasons for that, but probably one of the important things there is, as every model release increases the context window that those models are able to work on, and they are fine-tuning their techniques of what they use, the quality improvement from the model coming in is better, but it actually makes better use of all the context that we engineered around the model itself.
So, in a way, it's a self-reinforcing benefit of the work that we've done around the model, which is where our real value add comes in, which is our unique historical context, our 40 years of context graph of our data, and the engineering we're doing to bring in client context into that. Bringing that together is where that advantage comes in, which feels very, very real for us. In fact, the higher the quality of the model, it's making better use of the context around it, and that actually is translating into a much better quality-cost equation. But maybe, Kate, you can get more geeky in that.
Yeah, I'd say the lift there is quite consistent because of the fact that we have the ability of changing models behind the scenes. That abstraction layer that we have, as models and those providers, the frontier providers continue to improve, we test that and eval that against the different tasks that our clients are doing in our system, and we're able to make those changes behind the scenes to continue to give that consistent lift to the clients. So that's a little bit of the differentiation we have there from an individual frontier provider that prefers their own models. We have that abstraction layer to be able to pick and choose across the board, and help keep a consistent eval set up around what our clients are actually doing in the system.
Yeah, coming to your question, David, on comparing to other vertical harnesses. Look, every competitor has their own unique value proposition. What we believe other vertical harnesses cannot do is they don't have the access to, for example, portfolio analysis. They don't have access to internal research notes. We have over 11 million of those stored on our systems. Because we've been very conscious about client sovereignty for decades, we have that trust-based equation with our clients where they're happy to load their internal research, their internal positions. That's really, really important in building that higher quality context. We are very confident that it'll take any new provider, any vertical AI solution provider, it'll take them years, if not decades, to get to that same level of trust.
Can we go to Faiza over here.
Thank you. Faiza Alwy from Deutsche Bank. I wanted to ask about margins. I think, one, as you've talked about the upside, are there any potential margin implications of that? In the past when you've exceeded expectations, it's resulted in higher incentive comp and potentially lower margins. Just curious around how we should think about that. Secondly, historically, FactSet has had a bit of an inconsistent record on margins, and so would just love to hear how you're approaching that differently and maybe give us some of the building blocks around margin expansion over the next couple years.
Sure.
Sure. Yeah, Faiza, I'm happy to take that. Maybe a couple components to the answer to that question. The first is we expect margin improvement beginning this year, fiscal 2027, and into the future. That's based on what we see today. Our expectation is the operating leverage that we are wiring into the business is going to free up the capacity for both organic investment and margin improvement. It isn't a choice between the two, it's an and. The way I would think about our cost base is while we're going to continue to focus on growth, given our 95%+ ASV retention, thinking of that helps our business grow, that helps our business compound, that helps FactSet grow, and helps deliver value to our shareholders.
Being able to think about an operating footprint which is simpler, which is more scalable, where there is more and more resources allocated to best and highest use, that has been a journey that we have been on really for the last several months through to this year, and that has been one of the things we are focused on. All that is to say, I would not expect that trade-off to continue while we continue to pursue growth. We are focused on profitable growth. That is what will ultimately power our algorithm, and that is what you can expect from us.
Yeah, and maybe just to, you referenced the, maybe a shift in paradigm. We certainly are guiding, as we have done in the last few quarters, high confidence, right? We guide based on bottom-up modeling, where we see high confidence, and we aspire to outperform. We want to achieve the strategy, we want to accelerate AI adoption in the world, and we want to outperform both on growth and on margin.
Andrew.
Thank you. Andrew Nicholas with William Blair. Sanoke, in your presentation, you talked about how some of your biggest, most sophisticated clients are mixing and matching a little bit more. I guess a question there is just how much confidence do you have that that is incremental to consumption for and revenue for FactSet? And then somewhat relatedly, if you could just speak to the variation in how clients are interacting with you via AI, via MCP, via all the different channels that you are selling into. Is there a major distinction at this point between those biggest and most sophisticated and maybe the middle tier, and how you think about that evolving over the medium term?
Yeah. It is an ever-evolving landscape with our clients, and I would start by saying I would not equate biggest with most sophisticated. I think they may at times be an overlap, but it is not always true. And I would start by deconstructing what we do for our clients today, or have historically done, which is we have sold data, we have sold analytics, and we have sold Workstations which wrap both data and analytics. And when you think about the footprint that we have had with those clients, we have primarily been a distributor of the world's data and our analytics to the client. We have not really gone in and solved their internal data problems in the way we are starting to do with this new stack.
One of the largest growth opportunities for us is in this connected data space, and you will see some exciting demos later of how we are bringing our intelligent document service, our vectorization, our enrichment, our capabilities to clients' data. Now, that is a net new TAM that historically we did not really participate in, and that is an example where the decomposed or the deconstructed monolith, if you will, and the ability to distribute individual components and monetize them, we believe is a net value generator; it is a growth driver, and it is not just a substitution effect. In the range of clients, we see huge excitement in MCP activity, for sure, but we are starting to see great traction with our agentic solutions. The people are really connecting with them, trialing them, and we think the conversion rate of those into paid products will be very high.
It is still early days, but we see a plethora of new solutions being consumed. I do not think it is barbelling in any way. Even the largest, most sophisticated, and people who are building a lot internally are taking some of our full-stack solutions as well.
Let us move over here to Jeff.
Thanks. It is Jeff Silber with BMO Capital Markets. You talk a lot about pricing. I know at the last Investor Day, there was a lot of talk about moving away from seat-based pricing to enterprise-based pricing. Can you just remind us where you are on that journey, and now bringing in consumption-based pricing potentially? What percentage is it possible to quantify where you are now and where you think that will go over time?
Sure. I will start by reminding everybody that over the last couple of years, we have made good progress in reducing the focus on individual seat-linked contracts. Within the top 100 is where we track the transition to enterprise agreements the most, because there are large numbers of seats that might be linked into those contracts. As Goran shared earlier, we have made very good progress just in the last 12 months. Every one of the renewals that has come up, we have had success in converting it into enterprise platform agreements in over 40% of the time. That is continuing to reduce or de-risk us from any seat-based exposure. At the same time, any new contract we are writing is also constructed in a way that we have very high minimums, so we are effectively protected from the downside exposure on seats.
I would, going forward, expect us to make this not a relevant number in our overall numbers, because it is going to keep declining. When you said two years ago what was shared, I would say we have probably moved by 20 points from that down over the time, and it will continue to do that. So over time, this is not going to be a meaningful risk for us.
On the consumption side?
Consumption is early. I think maybe I will ask Goran to comment on how we think about consumption pricing in our contracts.
Sure. I think one thing to clarify or maybe to make clear is, the way we construct the base- level consumption for client is, we expect that that will satisfy their needs. We are not trying to really lower the barrier and then grow on top of that. I think that would be counterproductive to our relationships with our clients. But we expect some upside as the clients engage in more workflows, as they engage in agentic workflows, that we will achieve upside from consumption-based pricing there.
It is not just you can track it to our baseline. Our baseline does not contemplate any accelerated consumption pricing, so we think of that as upside.
C ome down front to Jason.
Thank you. Jason Haas from Wells Fargo. I am curious if you are seeing your customers start to use your AI tools to save on their own labor costs, and is that a big part of the pitch? Is it, "Hey, you should upgrade, you should move quickly on AI, and you will be able to get savings internally in your business"? I appreciate if you could comment on that.
It's too hard to tell because there are two things that are going on. There's lots and lots of new use cases that are popping up all the time at our clients, because as they get more comfortable using the tools, especially when there is a trusted tool, it's spawning lots and lots of new ideas. That's actually resulting in the labor that they have today actually getting more busy, not less busy. That's what we observe on the ground. The second thing is, they've added tons of talent in engineering, in data science. Bob can attest to this. Just before he joined us here from his other previous organizations, the increase in labor has been dramatic. A lot of technical talent has been added.
As we see today, we haven't yet seen the labor displacement TAM play through, at least in financial services, but it might very well happen in the future. Our proposition and our pitch to our clients is, you could at least save yourselves from having to hire thousands of very, very highly paid data scientists and engineers and rely on us to be an extension of their capability and use our high-quality solutions.
Down in the front here. Scott.
Thanks, guys. Scott Wurtzel from Wolfe Research. Goran, just on the go-to-market side, wondering if you can maybe talk about how you're thinking about balancing investment between regular sales headcount and these forward deployed experts, especially as AI demand continues to accelerate. It sounds like they're helping with the win rates. Just wondering how you're thinking about balancing those two sides of the investment scale.
I missed the first part of the question. The forward deployed experts again?
Yeah, just wondering how you're thinking about balancing investment between regular sales headcount and the forward deployed experts in your go-to-market approach.
We're gaining scale for forward deployed experts. I think both Josh and Bob pointed out, as we are deploying AI into our client service motion, we are realizing significant benefit from that. We are able to shift some of our client service, some of our consultants from answering client questions to more valuable work, like forward deployed experts. I don't expect a big investment in terms of, as we move to that forward deployed expert motion, we have 700+ consultants already in the field, and I think upskilling them and enabling them to interact with our clients to build new solutions is where we will gain that scale. I hope that answers it.
It's not a trade-off, Scott. At this point, we are getting both sides firing.
Yeah.
In fact, I would say sales productivity itself is starting to increase because of the use of agents and helping them prepare for meetings, doing RFP responses. Those 2,000 client engagements that Bob referred to on cybersecurity questionnaires, agents are helping us move those along a lot faster today than before.
Craig.
Thank you. Thank you for the presentation today. Craig Huber, Huber Research. Just a few modeling or financial questions, if I could. Walk me through the building blocks to get to the 6% growth rate here through 2029. I mean, how much of that is pricing? Is it like 2%-2.5%? I am also curious, you talked about your net employees count was down about 3% last year. How are you modeling, how are you thinking about headcount for the next three years in this AI world? Thank you.
So, maybe I will take those two. With regard to our revenue growth and pricing, fundamentally, we price to value, both value delivered, value created. So I would not think of it as a more traditional P times Q at the product level. What we are focused on is more, as Sanoke described, a client-led modeling, because we work from the client back. Our ability to deliver to clients is, in some respects, bounded by a client's willingness to adopt our own solutions and the like. So think of it as a value-driven model versus a CPI or any of that foundation. So that is on the pricing. Then with respect to headcount and mix, compensation-related expenses, non-compensation-related expenses, I would say, look, first principles, we are awash with ideas and awash with demands from our clients, which is an exciting place to be.
What we are focused on is trying to seek out profitable growth, and that profitable growth will involve allocating teams of people, our talent, to the best and highest use. It is that which gives us the confidence in the margin outlook that we provided. So I would not think of it as a straight headcount growth number or headcount reduction number. That is not the way we approach it. Again, think of us as operating from the client back, which is, I think, entirely consistent with the way FactSet has operated for the last 40+ years.
Let's stay up here with Curt.
Great. Thanks very much. Just two ones, if I could. Curt Nagle from BofA. Could we talk about wealth and just the opportunities? So, very high level, right, awash in opportunities in AI, but we didn't touch on the wealth segment. Do you see similar opportunities around MCPs, changing workflows, data? Then just kind of a broad question in how to think about token utilization, right? If agentic workflows are set to explode, you've got model optimization, maybe costs go down, right? You did describe it as an investment, so is it degradative, neutral? How is that factored into the model?
Yeah, let me start with the question on wealth. I mean, as you observed in the presentation this morning, we didn't go deep into our end markets. Really, that's just given the level of interest in AI and our transformation to be an AI solution provider. I think we focused the discussion on that. But I'd say, and if you stick around and watch one of the demos that we're going to kick off with, that'll give you a great understanding of how we are transforming the wealth advisor experience. How do they advise their clients with the full intricacy of having internal portfolio information, internal research from their CIO office, meshed with FactSet signals, research, market data? So that's a great example, and I would say that's what we're hearing at wealth across the board.
We're even seeing the starting point of clients wanting to work with our conversational agent and our mobile app products to distribute it to their self-directed customers. So that's a population of clients that historically our digital business has supported. We have 40 million customers that log into portals that we support every day. Now imagine the power of that with agents, right? So we're just at the start of this journey in wealth. We think there's lots of exciting opportunity there. Do you want to address the second part of the question?
Sure. Maybe I'll start, and Bob, feel free to chime in. Look, with respect to tokens, let me divide it in two. For our own usage, we're going to continue to invest in tokens. We view it as a resource. We think we're getting a lot done there. We think we're getting to a better output faster using tokens than not, right? Just creating a better, simpler, more scalable framework and the like. That's number one, and we're going to continue to invest in that. We're just getting more done, right? We're running a better software development, a better product development organization today than maybe we had been previously. So improvement there. Tokens will be a source of investment going forward.
I think your question, what you were really getting at was, what happens if client usage sort of runs away and there's this sort of crazy upside in terms of agentic usage? So maybe one statistic, and Goran mentioned it in his presentation that I might point you to, is 80 of our top 100 client agreements have some form of consumption-oriented pricing in them. We're not earning revenue, consumption-oriented revenue, on that today. That's sort of wired into the contracts. If that were to happen in that future state of the world, what the value equation starts to be is, again, we're aligned with our clients. We're delivering value based on the value we delivered, the value received, the value we're able to charge, and there's continued alignment of interests there going forward. So you'd see both, call it consumption-oriented pricing grow.
You'd also see token spend from FactSet grow. So if you put all that together, we believe that it's a good investment. We're very focused on, as Bob outlined, model routing, all the tooling, all the telemetry. One thing, maybe I'll commit to you, commit to everyone here, is one thing FactSet won't do is we're not going to give you sort of a gotcha token bill. We've been investing in the tooling, we've been investing in the monitoring to ensure that that's just sort of not a thing. We're not going to compromise decades-long relationships for a quarter.
Wahid.
Wahid Amin, Barclays. So, from the presentation, it seems as you've moved a lot of things pretty quickly. As you continue this transformation, what do you think you still have to prove to clients to continue progressing and continue meeting these goals for the next three years?
Look, I think, as I said, client adoption is going to be limited by the pace at which clients can trust the AI, and we are making strong progress with that. More importantly for themselves, they have to make progress in their own organizational adaptations, process transformation, to get the full value of it. That is going to keep the industry very, very busy, I believe, for the next several years. We see this having a lot of runway.
Alex.
Hello again. Alex Kramm, UBS. This is probably not a new question, and maybe some of this actually came up in the room earlier today, but I want to go back to the maybe the risk side a little bit, because you make a compelling case around where you have really sticky relationships, when people mesh, I think was the word, their data with your data, when they use multiple products, when they are the largest, most sophisticated clients using AI, you are definitely winning in the marketplace. But what about the tail, and maybe how has your view on the tail changed? Maybe Goran can talk about that too. You have, I think, 9,000 clients.
Yeah.
Some may be single-use case users that maybe continue to evaluate alternatives, and maybe AI changes those alternatives. So, how much time are you spending on that tail, and how much risk, maybe from an ASV percentage or so, do you actually see there?
Yeah, let me start, and Goran, you can go into the details. We actually find the, I mean, we do not really refer to it as a tail. It is just, we think of it as small, exciting clients with huge growth potential. We serve them in a comprehensive way. They may take one product line from us, but they are getting a lot of value out of it. If you saw the graph that Goran presented, the clients who are using anywhere from one to four solutions, that population, that is thousands of clients, that represented 50% of the ASV growth that we had last year. There is a lot of focus on it. It probably does not take the airtime in conversations like this, because people think of the largest and most sort of prevalent clients, but it is a very, very important part of the business.
But maybe, Goran, you can add more about how we cover that large population of clients.
Yeah, so, I actually think it's an opportunity. I think when we look at our retention rates, that retention rates at that end of the spectrum are not quite what we're getting with our larger clients. I think we are changing how we support them. I think that small, medium business client model, we're trying to really focus more on adoption so we can move them across that adoption line. I think if we can increase the retention rate there, I think we can realize some upside growth potential. I don't see them at risk necessarily from a vertical provider, so from AI in general. I actually think they, in general, they rely more on us for technology adoption than being self-reliant. I think we just, quite frankly, need to do a better job of retaining that long tail of clients.
Here's where AI is starting to help. It's still early days, but we have a really powerful tool that is giving a signal on client, what I would call trending of their sentiment relative to our product. We look at that every Friday, and we're looking at that sentiment graph and saying, "What can the salesperson in the field do, either digitally or physically, to actually improve that sentiment and continue to improve NPS?"
One last question from Craig.
Thank you. Just a follow-up. Very early in the presentation, you talked about a TAM, global market, roughly, what do you say, $40+ billion or something. You said, more importantly, you said it was growing 6%. Your outlook for your revenues next base case, obviously 6% through 2029. You did say that was a base case. Thought there was room for some upside here. How confident are you that you can actually outpace the market here in the next three years? I mean, it's the logical question. Thank you.
Thank you. It's a function of the fact that we serve a widely diversified client base that cuts across that whole market TAM that we talked about. We're serving 9,000 clients. They're across the entire industry. We believe we can. We have confidence that we are having very strong win rates. Our new strategy is allowing us to do things that our competitors traditionally have not been able to do. We have confidence in our strategy, but we're modeling based on what's already been observed and where we want to guide with high confidence the outcomes for the next, the outlook for the next three years. We believe that over time, our win rate will continue to improve. We'll gain market share, but we are not modeling it into the forecast. One last?
One last question in the middle.
Surinder Thind with Jefferies. Can you maybe talk about the level of investment relative to the margin expansion outlook, and how much of it is an arms race where the current spend levels are going to maybe continue, or maybe you actually even have to spend a bit more if competition tightens relative to the sustainability of that? Maybe as we get further into the AI cycle, can you start to realize more synergies, or does the pace of change dictate that you just have to kind of be in this perpetual spend cycle?
I'll start, and Kate or Bob can add. Number one, I'd say the heavy lift was what we've done in the last 12 months. We had to do a pretty significant transformation of the operating model to realize the full potential of AI, and that requires a lot of what we talked about, which is looking at the Workstation, looking at all our analytical infrastructure, looking at our data infrastructure, and working through what are the components that have even more value in an AI world, what are the ones that may not, how do we refactor everything and rebuild a brand new stack? So that's heavy investment. We are well on our way to realizing the value of that. I do think there will be lots of new synergies that come in as the AI cycle advances, because as I said earlier, model capabilities are not the constraint.
Now it's starting to come to, how do I get the best value out of it in a container that I trust, in a business system that I trust, with a service model that I trust? And here, there are lots of synergies in what we have in the assets and the capabilities we already have. I don't know, Bob, if you would like to add anything to that?
No, I guess what I would say is, and I covered a lot of it, is we're definitely creating opportunity and capacity there. As Josh said, I think there's opportunity for growth, and there's also opportunity for us to return that from a margin point of view, and I think we can continue to do both. I don't think it's an either/or, like you said.
Thank you. Well, thank you everybody for your attention, and for the questions. Really a lot of thoughtful, thought-provoking questions. I hope you got good answers to all of that. We have lunch now. That brings us to the formal end of the formal presentation. Over lunch, we are excited to show you some of our new products. Kristi Karnovsky and John Costigan, our Co-Heads of Product, are here in the room. They're going to introduce and showcase a number of the solutions. So if those of you who are sticking around, please go ahead and pick up your food outside, and let's get back in here in about 20 minutes, so we can kick off the demos, and I hope everybody sticks around. It's well worth the time. Thank you.