I'm Arvind Ramnani, AI Digital Platforms and IT Services Analyst at Truist Securities. Thanks for everyone sitting in. Along with me, I have CEO of Cognizant, Ravi Kumar. Ravi, thanks for taking the time.
Sure. Thank you.
Just quick introduction, and then we'll jump into questions. Ravi became the CEO of Cognizant at a really interesting juncture of the company and a crossroads of what was happening with IT services. Since taking over, he has made AI central to how Cognizant works with its clients, supported by over $1 billion investments in Gen AI. I think it's probably worthwhile mentioning, Ravi has been named 2x AI 100 list for the past two years. I think the reason this is particularly important is because that list had a 75% turnover. You just have 25 folks that have made it two years in a row and I think it's something to be proud of for yourself as well as Cognizant as well.
Thank you.
So look, we've had a number of conversations over the years. I have lots of questions here, not enough time, so we'll just jump right into the biggest debate for IT services. For decades, this has been a fairly, I would say, well-run industry. The industry had a very predictable path to growth. You add people, more billable hours. But now with AI, the opportunity basically moves very differently. What is already changing at Cognizant's delivery model because of AI?
Arvind, thank you for hosting me. Look, the first principles of the tech services industry have to be reforged because this is a very different tech disruption. Let me tell you what that reforging means and what we have done at Cognizant. The first and foremost, this is a business which is no longer going to be about human capital. It's going to be about human and digital labor coming together to deliver a tech-led transformation. Tech-led transformation for companies like ours was about building systems. In the 1990s, we built custom bespoke systems, then we became a system integrator for classical software. Then we actually became a services player for technology and software which was on the cloud and the plumbing was on the tap and the building was done by us. Fast-forward now, I think this is a technology which is action-oriented. It mimics human labor.
A lot of what we are seeing today is productivity-led. Still, we've not got to a point where we do new things in new ways. We're doing old things in new ways. The first and foremost, the first reforging of the first principles is we make the execution of this technology not just into systems but also business operations of companies. This market was $1 trillion before building systems at a time when enterprises were globalizing. This market now is $6 trillion because you actually have a total addressable spend around business operations of companies, and you do it embedded with digital labor. So that's the first reforging of the first principles of this industry. The second reforging of the first principles of this industry is you go more to an outcome-based business.
Enterprises across the world no longer want to buy software and no longer want to buy services separately. They want to bundle this and own. They don't want to own the agency of outcomes. They want to transition this to a provider who can do that. The third is it's going to be a platforms plus services business, which I believe is something we've been doing for years now. In 2017, we bought this company called TriZetto, and now we manage 200 million lives of insured healthcare in the United States. Platforms, human plus digital labor, outcomes-driven are the reforging of the first principles. If we reforge this, we see a bigger market than ever before. It's no longer a $3 trillion market, it's a $5 trillion market.
If you don't reforge it, you just end up doing software engineering, which then means you're doing it in the old way and you're going to deflate that model. But if you expand yourself into the business operations of companies, it's a 6x opportunity.
Yeah. No, that's a great big picture. But if you have to just get this down to specific customers, you obviously have a range of customers, somehow more, I would say, advanced in how they're using AI and how you're helping them. Can you talk about one of those customers and if you can name a specific name, or maybe the industry, I think that'll help me-
No, absolutely. So in 2024 and 2025, I would say it was the first chapter of AI, much more broad-based. It wasn't as nuanced. We dug a hole and a basement with the same machine. Most of it, the world is in a flattish growth trajectory. A lot of it was productivity-led. We kind of swapped wallet share to transfer the productivity to clients. If you could do that efficiently, you could actually win more wallet share with clients. It doesn't mean new spend cycles, existing spend cycles done more efficiently. That's what happened in the first wave. I would say later part of last year, we saw a second swim lane triggering off. That second swim lane was doing old things in new ways. Say, for example, for a large telecom client in Asia Pacific, we are migrating their mainframe into public cloud.
The idea of doing that is you could take the MIPS down, a line of code in COBOL used to cost $10 to refactor. Now it costs $1.50 to refactor. You transition that to public cloud, and the money actually gets refactored from a different spend pool to services as well as to cloud providers. It's actually old things being done in new ways. Recently, we set up a campaign for a frontier finance organization. In this swim lane, it's not new discretionary spend cycles, it's old spend cycles reorganized to give us momentum and a runway. Frontier finance is the ability to run your finance function using a frontier first model. Every Fortune 500 CFO spends 70% of their effort and time on controls and 20% to 30% on growth.
If you can flip it, you could actually orient finance towards growth, you could orient finance towards more real time, and you could power a finance function which is driven on growth imperatives rather than controls. Again, you take an old thing, it is done in a different way, and you do it in a new way. One of my favorite campaigns in recent times is pharmacovigilance or drug safety. I have four deals now where we have drug safety, which is called pharmacovigilance. We compete with clinical research organizations to win that business. We used to do a small sliver of work, which is mostly contact center driven. And now we could own the end-to-end cycle and deliver better insights, faster lifecycle, faster shrinking the speed at which you could do drug safety at a lower cost. This is a spend cycle our clients were already having.
They were not just having it with us. They were doing it with clinical research organizations. How do I make a function frontier-led and create a new spend cycle for us versus the clients are still spending in the same way? This is a swim lane which has got activated, and it is creating significant traction. You layer it with the consolidation, then you layer it on top of it with old things in new ways. And then doing new things in new ways, which kind of cater to growth imperatives of organizations. For example, for a wealth management firm in the U.S., we are building agentic work around their independent wealth advisors and actually deliver more capability to those wealth advisors who are actually on a subscription model with this company. So it is a new discretionary spend cycle.
Now, if I go industry by industry, Arvind, financial services is activated on all three swim lanes.
That is why for three quarters in a row, we have year-on-year growth, which is at double digit. Now, if all the industries activate this, you would actually see the industry back at double-digit growth. If somebody used to ask me a question, "You seem to be very optimistic. Why is not the industry growing at double digit?" The only reason why it is not growing at double digit is the first swim lane is deflationary. You actually consolidate. When you consolidate, you are passing on productivity to clients, which means it is going to be deflation to revenues. The second swim lane is new spend cycle coming to tech services. The third swim lane is discretionary-led, which is clients actually spending more because they are planning to do new things, and they are planning to do things which are related to growth using AI.
If all three are activated, you can layer this to double-digit growth as an industry.
Yeah. No, that makes sense. My own research actually suggests something a little bit different, which is some in the industry are not going to make it to the other side, right? Wouldn't you almost say someone like a Cognizant that's kind of leaned in fairly hard, right? You look at your commercial licenses with the frontier models. Anthropic, all your 350,000 employees have it. That's not the case across the industry.
Yeah
Could you make the case that when we see the final flip, everyone doesn't win, right? Everyone doesn't get picked up.
This is moving so fast. If anybody is telling you they've figured this out, they're not giving you the full picture. Here is my end state, which I can think of. The baseline is you need deployment capacity on the other end to embed frontier intelligence into everything you do in a company. Cognizant has 18,000 Claude-certified architects, but that's not my North Star. We are the largest on the planet. 18,000 Claude-certified architects, which is the highest level of certification in Anthropic. I'm just giving one example. In OpenAI, we have 6,000 Codex badges. That's the largest again in the world. In Gemini, we have 5,000. All of that is not my North Star. My North Star is bend the cost curve at scale and flip this model on its head and get the 300,000 plus employees in the company frontier-led. That's my North Star.
That's the baseline on which I'm running. That capacity is going to help me deliver frontier intelligence into enterprises. We are absolutely convinced the capability is out here, the production value is here. The gap between the capability and production value is the bridge companies like ours do this. It's a tale of two cities. On one side, we want to pace the frontier, which is our talk in the last one week, because the frontier is very dangerous. On the other side, we're dealing with the frontier can't do basic things because it's not grounded in the context of a company. I think you need that capability, and as long as you bend the cost curve at scale, companies like ours will stay relevant for the future. Go back to any disruption, including the digital wave. You had boutique digital capacity delivered at high cost.
I used to work for two such companies. They don't exist today. I used to work for a company called Cambridge Technology Partners, which was an absolute pioneer in digital technologies. It doesn't exist because it couldn't bend the cost curve at scale, which is what clients need, which is what enterprises need. The second important thing, as this technology evolves and as we learn through this process, between the intelligence and the enterprise, there are layers of value we can capture. Every time there's a disruption, that opportunity presents to us. We don't capture that layers. We always relegated ourselves to be a system integrator. We could capture those layers along with the deployment capacity at the end. At the end, the deployment capacity is like a services company.
But the layers of value are because this is a bespoke opportunity, we could build the trust layers, we could build the context layers, and we could also build the harness around it. I was telling Arvind offline, the harnesses built today, which are available today in the market, like Cursor and Cognition and all, they're for software engineering. The harnesses you could really build could be for business operations, because that's the universe. What can you do there? You could capture the context so that the next time you come in, the context is distilled enough, and therefore you're efficient in using your frontier. You could do the model routing so that you could straddle between the most expensive frontier to the cheapest open weight model.
You could coach the human with frontier intelligence live as they're doing the tasks so that they are not applying their mind on what actually transfers to the frontier and what remains with them. You could create a compounding factor at the end by leveraging the network of clients who deliver work specifically for things your clients don't mind sharing their alpha because it's not core to their business. Say, if you're doing accounts payable or you're doing procure-to-pay, it's not core to people's companies' businesses. It's not core to companies' alpha. So they would be willing to pass it on to get the network effect. If you're doing healthcare operations or you're doing underwriting, that is core to your business. You may not pass the alpha. Our endeavor from the intelligence to the enterprise, we have deployment capacity here, and we have layers of value.
Some we can capture, some we can try, some we will not be successful. There will be boutique AI ecosystem players who will come into the mix. There will be hyperscalers who will also absorb these layers. And some of the frontier model companies will also build commercial value proposition because they want to capture the value, because they don't have an ROI for the kind of investments they're making. Our endeavor is to not just hold that capacity at scale, bending the cost curve, but also capture some of those layers of value. Today, we are building a trust platform where we have agentic harness for business operations and for software engineering. I'm more optimistic about business operations. We are also trying to build one for physical AI, which is going to be the next wave.
We have built extraordinary muscle on context engineering, which is probably the new craft of writing code in businesses. If writing code was a craft, the new craft is embedding the context into intelligence so that intelligence is applied better to businesses. That's the journey we are going through, and therefore, I believe this business is going to be a platforms plus services business. How far have clients gone on the outcome pricing? I mean, that's an evolving science. Since 2023, since I've come on board, our time and material business, which used to be roughly 52%, and our fixed price plus outcome-based business used to be the balance. We have flipped that number completely. Right now, our time and material business is only 42%, and the rest of it, 58%, is fixed and outcome based.
Outcome based has a spectrum all the way from transaction-based pricing to outcomes. It's an interesting time because the first principles are getting reforged. It's also an interesting time because labor is not the only input to your business. Labor is one of the major inputs to your business, and frontier intelligence is embedded into it. How you can build the economics around it, how you build a craft around it, I think is very important because our clients are no longer buying in, or rather they have an intent to not buy intelligence directly. They want to buy it through us, which means we have to be efficient enough in embedding that intelligence and also routing it to the right economic models embedded into our work, and take the risk and manage the return through that process.
Yeah. No, that's a fairly comprehensive answer. Just a couple of threads on that. One is you said your fixed price is moved to 52%.
58%.
58%.
Fixed and outcome based.
Yeah. And within that's level you said that contains many different segments. But the segment that's basically tied to AI, whether it's the 18,000 Claude-certified engineers or the 6,000 Codex engineers, the folks who are basically trained and who are extensively using AI tools, is that bill rate like a step function different or just like a slight premium?
I mean, it's interesting. On one side, we have outcome-based pricing, and I put this chart of cost observability on the Y-axis and outcome observability on the X-axis, and you always think the upper right is the best one. The upper right is not the best one here. Clients actually look for the best suitable candidates for outcome-based pricing are high outcome observability and low cost observability. Interestingly, if the observability on cost is very high, clients don't want to do outcomes because they can see the cost. So you could do an outcome-based pricing, but you're not going to get a nonlinear return. If the cost is not observable, clients want to leverage that. So interestingly, outcomes have to be on this side.
Observability on cost has to be lower for clients to actually say, "I want to do outcome-based pricing." If the observability is high on both sides, you will do a fixed price deal, and fixed price deals are not outcome based. Sometimes fixed price deals can be less margin accretive because the observability of cost is so high that clients will not pay you more than what you should, and you still take the risk. So outcome-based pricing starts from a spectrum of managed services to transaction-based pricing to owning the outcomes.
Owning the outcomes means, say, for example, in our TriZetto business, we went from selling perpetual license to subscription-based license to transaction-based pricing to now our clients are saying, "Why should I pay you for transaction if my transaction is auto adjudicated with AI? I should be paying you on number of lives you manage for us." Number of lives managed to us is actually on the other side of the spectrum, which then means I am actually managing number of lives versus number of transactions underneath it.
It is a spectrum all the way, and getting there is going to be a long haul, but we have spoken about outcome-based pricing for 25 years. This is the first time we have got a chance to own it. Because we are doing operational work more than software engineering, we could own it. I could go to a company and say, "I will own your know-your-customer's process," or, "I could own your accounts payable process and give me a cut of it." A good example, we work with one of the largest food distributors where we manage their agentic process of credits.
Food distribution has a low margin, and the credit cycles, when you return your goods back because you do not like them or they have not come to your expectation, the money flows back to you. The goods go back, and the money flows back. It is a long cycle. We could shrink it and increase your working capital efficiency, and if we do so, we can take a cut out of it. Clients have not gotten that far. They still engage with us on a fixed price model. Over a period of time, that would happen. Now, coming to time and material, interestingly, that is becoming outcome based as well. I have an AI-infused rate card , which I introduced last year from A0 to A4. A0 being all human effort, A1 being human effort validated by machines, A2 being machine effort validated by humans, A3 being all autonomous effort.
We presented it to a few clients. The clients loved it. They gave us a premium on the rates as you go from A0 to A1 on the human labor. Then our clients came back and they said, "Wait a minute. I thought I got a great deal on this, but the inference and the pre-training costs are on the tap, and that is not controlled. I am paying directly. If I put all of it together, I do not think I am getting a good deal." Now clients are saying, "Take the inference cost, take the pre-training cost, embed it into your billing rates, and tell me what your new billing rate is." Which means I have to give a billing rate for human and digital labor.
If I have to do that, then I have to have traceability of the tasks which these people are going to do so that I could measure the pre-training and inference costs. It is a craft we have to build so that then becomes the opportunity to arbitrage with clients on expertise. It is not yet happened, but clients are starting to think about. The A0 to A4 rate cards have happened in many of my clients. You do not give the rate card based on experience, you give rate card based on whether it is fully autonomous or fully manual. But embedding digital labor is one more level of complexity, which means the tap on Anthropic and the tap on OpenAI is going to be through us. By the way, we have a pre-buy arrangement with all the three frontier model companies.
There are deals where we have told our clients we will deliver this outcome. We went public on one particular one called Travelport, where the client is actually delivering operations and software engineering embedded with a frontier model. They are not paying the frontier model company, we are paying the frontier model company. We have started to do that kind of work, but it is a craft we have to master so that our clients give us that work along with the frontier intelligence because we can do it better than them.
Yeah. If we can just continue on that thought, where basically clients can very well go and buy these directly from OpenAI and Anthropic. They know how to reach them. What's the value proposition or what is your value proposition as Cognizant to say, "Well, if you buy from us, then
Yeah. Essentially you're not buying capacity from us.
Right.
You're buying an outcome. You're buying to deliver a task. We are figuring out how much of digital labor, how much of human labor, and if the digital labor should come from an open weight model, come from a cheap frontier, come from an expensive frontier, which means they are not actually managing that risk and they're managing the outcome. So don't come to us because we have wholesale capacity on Anthropic. They can go directly to them.
Yep.
Come to us because we know which frontier to use to drill the hole and which frontier to use to drill the basement. We know where to use a frontier and where not to use a frontier.
Allow us to manage that and focus on the task accomplished. That's the thesis, and that's why we have pre-bought this capacity so that we can deliver to that outcome and deliver to those tasks. Software engineering is relatively easy because it's very mature. Business operations is not easy because it is evolving. I would like to do it because I want to master the craft. We have to pick and choose what business operations make sense for us. Healthcare operations make sense for us. I'll give you a good way to say which flows are more amenable for agentic. I call them bounded workflows. Bounded workflows have four principles: structured inputs, measurable outcomes, high transactions, short feedback loops. If it ticks all these four boxes, I think I can embed intelligence, measure it well, deliver to an outcome, and take the end-to-end responsibility for it.
That is what we are looking at. We are looking at healthcare operations, billing systems. These are high transactions, structured inputs, measurable outcomes, short feedback loops. Mortgage operations. These are solid opportunities. On the horizontal side, legal operations, financial operations, or the CFO's function or the frontier finance function, these are solid opportunities. That's how we are at least learning through this process.
Yeah. Terrific. Shifting gears a little bit. You'll have fairly, I would say, 360 or comprehensive relationships with Anthropic, OpenAI, right? On one hand, you'll have the largest or the second largest footprint of Anthropic, 350,000 licenses. I think maybe Deloitte signed one, but I don't think they're fully ramped. As of now, you're probably the largest footprints. On the other hand, you have these folks who are Claude-certified engineers and basically experts. Maybe there's a smaller set, but fairly advanced in what they're doing.
Yeah
At the same time now you have some customers who are also wanting to buy, so you are generating a lot of revenue through them. And you have this going on where you are essentially having more of a customer type two relationship, and then at the same time, Anthropic is out there basically building their own deployment layer. And clearly the classic frenemy sort of situation. Where do you think this relationship goes in five years?
Look, I was at one of these conferences where somebody from a frontier model company asked me that question, saying, "Are we going to compete with you at deployment capacity?" I actually said, "Absolutely no." On the contrary, you are making the world believe that deployment capacity is needed. On the contrary, you are making the world believe that this deployment capacity is needed and so much value, the capability is here, the production value is here. As this progresses, bending the cost curve at scale on deployment capacity will be tested, and we will be on the front of that. So having the boutique capacity, frontier capacity from these companies, including private equity companies which have floated these joint ventures, I think is a great thing because it just fuels the need for services.
Remember, everybody put money into the intelligence, and then they figured out there was very little economics to put value in the layers before it hits the organization. So they couldn't ground the technology in the hustle and the heterogeneity of an enterprise, so they couldn't get value. So now that the intelligence is more ROI-driven, that value is drifting into the trust layer, it is drifting into the context layer. It is drifting into reinvention of the process. Reinvention of the process is real. If you are not reinventing the process, you are not reinventing the operating model, there is no way you are going to get the ROI from this.
So the more this happens, the more I believe the need for frontier capacity at scale, at a lower cost point is going to come into picture, and then companies like ours will have more mainstream role to play in this new transformation. So this is already happening now. Everybody is backtracking on this first chapter where they did the science experiments. They invested heavily on generic applicability of AI to now specificity, more ROI-driven, more economics, and all of that is now coming into picture. So I actually feel it is a great thing. It just gives endorsement to important constituent in that layer stack, which is deployment capacity.
Yeah. Terrific. I do want to ask one sort of prickly question that everyone's been talking about over the weekend where essentially there's a chance that AI ends up killing us all. People assign like a 10% chance. I think last year, what we heard from Anthropic and OpenAI is like, "Oh, I'm kind of afraid that in six, eight months there's going to be massive job destruction." Now it's going to be like, well, they backtrack from that. Now it's like, okay, now maybe AI will just kill humans itself. What's your sort of perspective on that?
My own take on this, you don't get into the ring and then say, "I can't take this to the finish line because it's going to be ruthless." The reality is pacing the frontier is what this whole thing triggered off-
Of course
-With a bunch of employees writing, and then Dario putting this essay up, and depending on who you talk to, they would tell you to pace the frontier or not pace the frontier. The reality is the duty of care is your call. You own the frontier, so you should decide whether you want to pace it or not pace it. If you don't pace it could be dangerous. If it is dangerous, the economic consequences of the liability you carry is going to be yours. If you pace it, somebody else is going to advance it, and the world is not in a place where everybody is going to come and there's going to be a commonality of purpose, and there's going to be an agreement of what the pacing should be and what you're going to slow down on.
If I have a genie in my pocket, I am going to say nobody else should have one. I do not think we have the option to take a call on pacing the frontier or not. If you make the duty of care as a responsible and a trustworthy duty of care, then your clients will come and buy from you because that is your duty of care, that is your calling card, then everybody else will actually put that calling card, and the industry will already baseline itself with its guardrails and pace it in the way that they can control the guardrails. This is a general purpose technology. It will improve over time, and it will spawn downstream innovation.
It is up to you to improve and control rather than saying, "I want to control it." Who else in the world is going to agree to control when you have advanced and the others are yet to? And it is in the hands of a few players versus an ecosystem which has access to it, and therefore, it is a level playing field for everybody else to decide that pacing the frontier is the right thing. Of course, there should be regulatory frameworks attached to it, but I do not think the players decide everybody should pace. If you want to pace it, you should pace it yourself, and you should be accountable for. Do the balance between capability and responsibility, and predictability and reliability. You have to set the balance for yourself. You have to set off the capability to the side.
Now, the ones who do not have the capability are going to run the extra mile to get to that capability, and the ones who have the capability will be worried whether they are going to be accountable to the liabilities of the consequences of what the danger of this technology could be. I think this is a debate you cannot win. You could argue the debate on both sides. There has to be regulation for sure. But it is the same set of companies who actually painted the doomsday situation. In the U.S., the anxiety on this technology is higher than the excitement. In China, the excitement is higher than the anxiety.
We created this situation, so we have to come out of it anchoring this to more jobs, anchoring this to what AI can productively do, like cancer care and cancer cure and unlocking material sciences and stuff like that, and curing disease. Those are extraordinary north stars to deal with. But it is our opportunity to figure out how do you advance and control, how do you advance and control so that you get the best out of this technology.
Perfect. Just a little bit over time, but just one final question. If you look ahead over the next year, what's the one sort of metric investors should be most excited to be tracking?
Any metric you take. First of all, the most important metric is growth.
Of course.
If the industry was a double-digit growth, nobody would be asking me, "Tell me the proxy for growth." The industry is not a double-digit growth. Financial services, at least for Cognizant, is a double-digit growth for three quarters in a row. AI revenues, AI projects, frontier capacity, all these are, I would say, not baseline metrics because everybody can paint everything as AI. Everybody can say everybody is frontier capacity. So it's hard for investors to believe in those metrics. Now, if I were to go back to the principles, which is human plus digital labor, platform play, outcome-based play, the two metrics which will be tangible and measurable and baseline for everybody is how much percentage growth you have had on revenue per person and margin per person.
I think that is a reasonable metric to measure, and hopefully, the industry goes back to double-digit growth at some point of time, and you are kind of anchoring this as the journey to get to double-digit growth.
Terrific. I have another 10 more questions, but we are a little bit over time. But thank you so much for taking-
Thank you. Thank you for the opportunity, and thank you for a very thoughtful conversation.
Of course. Thanks. Thanks everyone for listening in.