Cognizant Technology Solutions Corporation ($CTSH)

[Audio Gap]

Starting to see unlock of vulnerability is security-led with Mythos or 5.5 GPT, where you could do vulnerability discovery at machine speed, but you can equally remediate, re-factor and patch at machine speed. And that opportunity is almost like a Y2K moment for us because as you open the lid, you don't just look at one thing. You look at [ bit and core ], you look at design deficiencies. You do a whole lot of re-factoring. So the ability to take your old stuff, pull in new ways to do it, underwrite that capital for the new things. Now new things which I'm talking about is a completely new terrain, which is primarily applying software on operations of companies. Now historically, portions of those operations, which were deterministic in nature, were done by classical software. Portions of the things which were not deterministic, which had judgment, which had -- which could not be codified, we put it into human effort. Just an example, the health care platforms in Cognizant have a flow-through of $500 billion of spend flowing through our platforms. We do health care operations for our clients. Now I could [ agentify ] another $4 trillion work, which was administrative in nature -- remember, health care has 20 million people working, only 5 million to 6 million people do care, 14 million to 15 million people do administrative work. I could agentify it. Now that doesn't made big discretionary spend. You just have to front-load it. You have to transition human labor to digital labor. It is self-funded. So you could take the steps even in an uncertain situation -- this is what I do with clients, yesterday I side was in my health care conference, we had 1,000 clients there. And the ability to take some of that and put it back into care I think is a phenomenal example. So in summary, the framing I have in my mind is the AI capability is absolutely moving at a rapid pace. The bridge to production value has a big gap. The more the capability is the more the production value gap is. And what does production value? This is a contextual technology. Production value means you have to ground the technology into what clients wanted work to be transferred to AI, which is workflows, controls, guardrails, trust layers, evals, context and the tribal knowledge, all of that put together, the more the technology is evolving, the more the technology is advancing, the more is the gap to enterprise value. Not for software engineering. Software engineering is already mainstream. I'm talking about operations of companies. I mean, if $1 trillion were spent in the last 1 year on infrastructure build-out, another $2 trillion to $3 trillion is going to be spent next 2 to 3 years, all of that buildout will only stack up if you are able to apply it to operations of companies. And the bridge to enterprise value, the bridge to enterprise production value, I think, is a big gap. So clients are asking me, how do we actually bridge that gap? So that conversation is happening now. So therefore, I'm actually much more optimistic about how AI is a tailwind to our industry. You need a ton of work to be done to get to that production value. And the production value will come from agentifying finance operations, legal operations, HR operations, customer service, and vertical operations like health care, mortgage operations, a whole bunch of things, which was a human endeavor. It's a different topic on a different day, to talk about what will be the next endeavor. But all of that, we have a struggle between -- interoperability between human and digital labor and building those flows in companies.

Yes. So I like the GAAP discussion, being on the bridge, and I think AI being a tailwind, I think these are all important messages. But investors...

More the capability, more the tailwind.

So investors are, of course, looking at the financials and looking at the performance, right? And the news cycle is so intense, Ravi, and to translate, right, what's cyclical versus secular? I'm curious just to get it out of the way. I know you just set your outlook just, what, 3 weeks ago, and there's been so much change since -- and we'll talk about some of it. Has that changed your confidence in the outlook and what you see in the near term here?

I mean look, 2 to 3 weeks is too less to really change the paradigm. But the one thing I can certainly tell you, everything I said so far is no longer conversations yet to happen. The conversation is happening now. I already have a large mainframe deal where I'm using Claude and transitioning work to AWS Cloud. I have at least 5 to 6 SAP migrations happening. I have 10 to 12 discoveries happening on Mythos and 5.5 GPT. So those conversations are accelerating. I would say, if you go back to the last 2 years, the industry was layering consolidation an expansive spend related to consolidation as a growth lever. That is right at the base I'm layering in these new value pools of old things done in new ways, which is unlocking mainframes, unlocking SAP migrations, unlocking vulnerabilities on landscapes and layering it further. Once I lay then the spend on the operations, which is also self-funded, it doesn't need so much discretionary because you are actually eliminating work versus creating more work. Once the unlock on growth happens, you are going to use the same set of people to create more throughput, you're going to see new products and new services powered by AI. So if you put those -- if you put that layering in, you'll see an inflection point at some point of time. So this is not about -- my framing last year was, what does AI do and what does humans and tech services company do? My framing has completely changed. My framing now is the capability is right up there and it is further going up. The production value is here, and the gap is huge. That gap needs a bridge. And that bridge will come from companies like us. If you believe that the capability will not actually impact operations, then you almost have to question why the infrastructure build needs to be that much. So that's the way I'm framing it. I mean it's a question of time, does inflection point has to happen? I mean Financial Services for Cognizant is already a double-digit growth. My BPO operations, which is where the expensive opportunity is, we have been on double digit for 2 years in a row. Infrastructure services, because there's a sprawl of infrastructure, that is close to double-digit growth. Software engineering is deflationary. The expansion attached to software engineering is taking off now. So I mean, we just have to keep layering this and we'll be back to industry-leading growth. I mean we're already on the winners circle, and I'm less interested anymore about being on the top of the heap. I just want to take this to breakaway growth. And that is my new aspiration. I mean being on the top of the heap is no longer good if you're on the top of the heap. You should actually look for breakaway growth.

Yes. No, I'm glad you said that. And look, getting back to the winner circle, I know it was not easy, and that was a lot of you were doing from an execution standpoint. So you should be credited for that. But you're right, focusing on that is missing the bigger picture. And I'm asking the reason why I asked, right, because the new cycle is changing so much. We're going to hear from OpenAI tomorrow. And everyone is asking me to ask you, so I'm going to ask it here, right? With OpenAI and their DeployCo, and then Anthropic similarly working with these investment partners to build out their own deployment or services capability. What does that mean, right? You're talking about this group that [ were focused on that ]. Was that the change in the competitive landscape in your mind? Or is it more of the same?

It doesn't. Actually, on the contrary, it reinforces the fact that there is a gap between production value and the capability of these models. We underestimate the heterogeneity of enterprise, underestimate the fact that this is a contextual science. Everything which was supposed to be deterministic has already been in classical software. This is deterministic. It has to be grounded. It has to be harnessed. So on the contrary, it reinforces the fact that we need that bridge. You can argue whether that bridge will be companies like us or it will be new companies coming into picture. I don't think those deployment companies have been built for scale. I don't think it is built to make -- monetize the bridge. It has been built to get access to the distribution network, which is needed. And that distribution access always was the reason when new technology came into picture. When enterprise software came into picture, we had Oracle Consulting in the mix. When SaaS software came into picture, we had professional services supporting SaaS software. Cloud -- the embrace of cloud happened where professional services coming into picture. But it never took off at scale to build what is needed for the Global 2000. So I see this as an endorsement of what we do. I see this as an endorsement that this bridge is needed. Do we need to do it differently? Yes. I mean everybody is now talking about a forward deployment engineer. Forward deployment engineer was the perfect thing in the last few years, when the world view was everything was broken in an enterprise, you have to bring everything together, you go with the machine and you go with this forward engineer and you tie it together. The world view is no longer about everything is broken. The world view is you want to take this machine, which has digital labor, along with you, and you want it to be integrated and you want it to deliver an outcome. So we have to reforge our first principles. Our first principles have to be reforged where we go from a system integrator to an AI builder. I've mentioned this. We go from a pyramid of talent to integrated interdisciplinary talent with human and digital labor. We go from a services company to a platform as a services company. We go from managing project outcomes to managing operational outcomes for enterprises. So the roles we need now is frontier engineers and frontier operators. Frontier engineers being people who can engineer equivalent to, in a way of forward engineer, the machine at that time was [ ontology ] with machine learning. Now the machine is generative AI. You also need frontier operators. What I mean by frontier operators are, if clients are going to run operations with us, some clients are actually going to say, you run those operations for me with human and digital labor. And that is not what frontier engineer is about. Frontier engineers engineer this product. I'll give you an example. I have 5 banks who are doing know your customer with me. They don't want to outsource that function. What they're essentially seeing is we love what you're saying on a agentic, we want to agentify giving you a fixed price proposal for agentifying your customer. There's another set of clients who come and told me, here are the stuff which you do in health care today on your [indiscernible] platform, there is a ton of labor sitting outside. It could be codified. Please identify, run those operations for me. So I need a BPO professional who can work with digital labor and have an integrated digital and human labor actually deliver an outcome. That is the new age BPO we are actually building, and my BPO business is actually a double-digit growth. So you need frontier engineers for things which you want to engineer for your clients, agentic. You need frontier operators for things which you want to actually own the outcomes, operational outcomes. So with TriZetto, we are going from -- we went from perpetual license to subscription license to BPaaS. Now we are going -- we have started to work on per member per month to manage lives. It doesn't matter what transactions. In fact, in an ideal world, you shouldn't have transactions, you shouldn't have claims. It should be a claimless world. So just manage per person per month, here is the menu card. So we have to be a platforms company. Either we have to build our platforms or we have to work with third-party SaaS companies and integrate their platforms into our services. Because clients are neither going to look for SaaS software on its own in some places, neither are they going to look for service, they don't look for AI outcomes. So what else has changed in the last few weeks? Tokenization has become a huge thesis for me. And I'll tell you, in my earnings, I spoke about AI-enabled rate card, A0 being human effort, A1 being human effort validated by machines, A2 being machine effort validated by humans, A3 being autonomous. And now clients are saying as you go from A0 to A3, the premium on the rates will go up, but the number of units will go down, because there's machines attached to it. Now clients have started to say, wait a minute, you want to take the human effort accountability, I'm going to take the machine effort, and I don't know how to do this. So take this over and run, which then means I should build a hardness around tokenization. Why? Because I could do it with hundreds of clients so I should be able to do the same things in a compounding way. That's why clients came to us. Clients said, "Oh, you do this work well. You've done it with 50 clients. Give me the people who have done this before, show me references, I'll give you the business." That's how clients did. Now tokenization could be a new function value for IT services. So we have metered inside the company projects, people and the kind of work we do. That metering leaves an institutional hardness. Earlier, it was at the minds of people and repeatable artifacts. Now it is in the harness. I should be able to deliver your digital labor more reliably, cheaper, cheaper is important, till token costs are set up, and more predictably, which then means I take a high risk on tokenization, higher risk on inference costs on tokens, if I'm doing operations. If I'm doing development work, the token costs are less risky. If I'm doing inference work, I should be able to. So we have now started to build a craft on tokenization. And this is a discussion many of my clients are doing now. They're saying my bills are going up, would you be able to take this over? We use these tokens to process invoices or process claims. For each of them, we have a way to spec it, we have a way to size it, and we have a way to price it. So effectively, that's a new moat. And it will come from the community knowledge of clients because clients actually -- if you do it at 30 places or 50 places or 100 places, you'll be able to do it better than your clients. So that's the [ craft ] we are building. Is it completely efficient? Not yet. Am I going to falter? Maybe yes. But I'm going to build a moat, which will help the tokenization process to be more scientific versus today you trying to measure tokenization based on consumption versus measuring on value.

This tokenomics thing, thank you for going through all of that, right? But we get this question a lot, right? I mean creating a new model and then pricing a new model and you're going back and forth with the clients and you're learning -- I appreciate that you have a lot of scale and you're accumulating all of these focus, you can price it differently than consuming it individually. But what's the risk of something going wrong, Ravi, in terms of just like thinking back to the fixed cost?

Absolutely. Absolutely. In software engineering, the risk is low because...

[indiscernible] with that.

Also, you are developing something, you're not running it. The cost of tokenization is a higher risk in inference and lower risk in developing. If you're running operations of companies, you have to be precise on the tasks you want to do for a client and you should be able to spec the job. So you have to pick your swim lanes where you want to. I mean you could be blind-spotted by things which have high influence costs. So I'm not going to run wild of this model. I want to run in a controlled way to figure out which swim lanes we can precisely size it, and which swim lanes will have throughput so that we can build patterns and, therefore, the compounding effect. I mean the compounding effect is similar to autonomous cars. Autonomous cars drive better than you not because they have your learning. They do better than you because they have community learning. So wherever there's a repeatable -- rinse and repeat template, you should be able to do it. Now is this going to be a long-lasting moat? I don't know. If the cost of tokens is so low that it's like a utility, then cost is not going to be a driver. It has to be driven by velocity, and because it's a contextual science, it's also driven by reliability and predictability.

So adoption wise, how quickly do you think you'll learn and be able to evolve and commercialize this model? Is it months? Is it quarters? When should we be asking you the right questions on, hey, you figured it out? Forget about the moat, more the adoption.

I would say the ones which have been historically outsourced, like health care operations, mortgage operations, customer service, F&A functions, are easy or relatively easy. The ones which have been historically not done, like legal operations have not been outsourced to companies like us, HR operations have not been -- that will be harder. The vertical ones are a longer runway. The horizontal ones have a lesser opportunity, if I may. Because the vertical ones actually have the most administrative labor which is -- which can be actually shrunk. I mean I told you about health care just as an example, 20 million people work in health care, 5 million to 6 million people really do care, 14 million, 15 million people do other things. And it's a high-churn industry, so it's not like employability at -- on a sustainable basis. So I think we have to pick the right ones. Health care is certainly one I'm picking. In fact, I want to make sure that the hustle is around TriZetto. So the rules engine of TriZetto, we want that to be the real moat from which agents are accessing rules and guardrails and everything else. So we make sure that that is where it is. If I have to pick a few more, Tien-Tsin, just at a high level, wherever classical software did it in a less effective way, like billing systems, recordkeeping systems in life insurance or mortgage operations. I mean these are the places. These are not the best-looking things in classical software; they potentially become the best-looking things for AI.

Okay. So thinking as we maybe hopefully get you back here next year, thinking about this is developing and your reforged principles, how ahead of the curve do you think you are on this, Ravi? Because it feels like there's some margin accretion potential and usual financial questions we ask. But just thinking about how far ahead you are versus the peer group, how confident do you feel around that? .

Look, I'm layering in consolidation of work, which is giving me some runway. I'm layering in these new value pools of all things done in new ways, which is giving me some additional incremental growth. Now I'm layering in new things in new ways. New things in new ways is also operations, self-funded. Once we see the macro improving, you will see new products, new services coming in. I think last time I spoke to you, we spoke about a digital nurse, which is a new thing. It's not a labor pool existing, a digital nurse for people who are doing dialysis treatments in their homes. We built something like that, we're building something on wealth management. So we're building a new -- those have not fully taken off except financial services because of discretionary. So that layering will certainly happen. So in a year is what you're saying, right? In a year from now, I think what you have to judge companies in this sector, including us, is are we -- I mean, it's hard to say what is AI revenues and what is not AI revenue. So it's hard because there's no real baseline. Revenue per person, margin per person is an important metric. And I would say revenue per person and margin per person because I've changed my pyramid mix with our Leap program, what I'm really saying is the pyramid is broader and the pyramid is shorter. So I'm delayering the nodes where there is administrative work, and I'm only having player-coaches in the middle. And I'm hiring significantly more number of people at the bottom, higher than last year. And last year we hired higher than the year before. So that whole thing will give me margin per person higher, but it will not give me a revenue per person higher because I'm broadening that. So I want to create the right balance. Right now, we've got a runway on revenue per person, we've got a runway on margin per person. Margin per person has got a higher bump than the revenue per person, in my case, because we're broadening the pyramid. But both should get a bump. That's my sense. Now what we have achieved so far, is it good enough? It is -- I would say, it can be better. Revenue per person and margin per person in the TTM basis has just about got to double digits, so we have to expand that. The platform play will allow us to increase the revenue per person and the margin per person because platforms are an important part of my thesis. We have set up a new platforms group dedicated. Some we want to build, some we want to buy. Some we are going to invest. We put up a venture arm, we announced a few weeks ago, and some we want to partner with third parties. I think the throughput from the 3 frontier model companies: Anthropic, OpenAI and Gemini, all 3 are going to be super important because that's the machine we are carrying now. I think you should also -- I don't know how to judge this on a common scale, but the number of frontier engineers, the number of frontier operators we've been developing, I mean we are a company and we are an industry which has not built human capital based on buying stuff. We've built it based on building it. And this is a new -- we are getting to a new era where a new set of companies and a new kind of people who will express in those companies will work, and those new set of people are people who can actually take these frontier models and deliver operations for companies. So I am measuring it on certifications, which is an external metric. That is something I'm very keen to pursue. And I think for our scale, will be super -- it -- that's the kind of capacity needed to bridge capability to production value at scale, at much better economics, at much better economics. So I mean if you put all of this together, you should start to -- we should start to go back. I don't know when, but we should start to go back to growth rates which make the industry attractive.

Yes. No, the plan makes sense and it's going to be fun to track for sure. I have to ask one follow-up to that though. Thinking about the capital intensity of doing some of this, I know you're acquiring [ Astrea ] and getting into managed services infrastructure. So getting to where you want to be, Ravi, is there a need to do more inorganic to get to the right place? How do you balance that?

So we did a buyback for the morning, for $1 billion. We increased it by $1 billion. We had $1 billion of buyback in January. We felt like that was a great way to show confidence and great way to back our story. I think the M&A we do should fall into this thesis of platforms are operations. And we'll be very purposeful in doing that. Astrea has not closed yet, but one of the reasons why Astrea made phenomenal sense to us is the infrastructure sprawl, which is going to happen, we are already close to double-digit growth in infrastructure, we felt it is an extraordinary opportunity for workplace services, data services, data center services and network services. That's what the company does. And it doesn't do it on human effort; it does it on a user basis. So they do it on a per user basis and they do autonomous infrastructure in workplaces. So it's a complementary capability, platform-led, and a new outcome-based model because it is not based on human effort, it is based on the number of users. So it fell into our thesis, and that's what we did. I mean would we be opportunistic in places where we have lower presence? For example, I have less presence in Asia Pacific, I have less presence in Europe. Of course, we'll do it. But my general strategic intent on M&A is to layer platforms and operations, because that's where the bridge to enterprise value, the bridge to enterprise production value is very high. So that's been -- that's how I look at M&A.

Okay. That makes sense. So we're just about out of time, Ravi, I wish we had more. But just thinking about piecing this all together and, hopefully, we'll have you back next year and I'll ask it again, but what outcomes do you think will be most obvious that you're pushing towards that will come back and be like, "Oh, wow, your progress and your reforge principles are working?"

Certainly, revenue per person, margin per person is one metric to watch. I would say more fixed price, more outcome-based pricing. We have 50% fixed price, 10% outcome-based roughly. We moved -- we flipped the whole thing. We were -- we almost added 10% into fixed price in the last 3 years from 2023 to now. Fixed price because the engineering on AI where people are not willing to give you the operations will potentially go fixed price. By the way, I have to make that distinction there that the AI-infused rate cards is also new stuff. It might show up on time and material. But it is new stuff because I've been fusing digital labor and I've put this new stack. And we are the only one who are actually propagating this to our clients. But that will be another metric to say we are progressing that way. The work we do with our platforms group is a very important change in our reforge first principles. The work we can do on outcomes is a combination of fixed price and transaction-based outcome-based pricing. These are the reforge principles on which you should see whether we're progressing in the right direction with the right velocity.

Okay. Good. No, the fixed price piece, I think, is important. No, I appreciate you going through all of that. You're pushing through a lot of change, which I know isn't easy and is forcing us to study. We need to come up with a way to phrase forward operators, we need an acronym for that. But I appreciate all this, Ravi.Thank you so much.

Thank you. Thank you so much for hosting me.