Unlocking AI Value: Salil Parekh on ROI, Jobs, Data, and Governance

Rachel Oppenheim:

Salil as CEO of Infosys, you employ hundreds of thousands of people across the world. That's intimidating and are operating across more than 50 countries. And I know you're partnering with many of the world's kind of largest enterprises who sit across geographies, who sit across sectors are navigating how to adopt and employ artificial intelligence in a meaningful way. It's great to have you with us, very excited for this conversation. So to begin, here's my first question. Traditional ROI frameworks were built for capital investments that were very predictable in terms of when their payback periods were going to be and one of the things that I think a lot of CEOs across this week have been discussing is that the kind of the picture with AI, the value that's delivered, it's also a bit diffuse, it's a bit more cumulative, and it's really hard to measure and to attribute. So I wanted to really start there and just ask how should the CEOs that are here at Semafor World Economy and how should their boards who are increasingly holding them accountable both to AI implementation but also AI cost discipline, how should they think about these kind of competing and converging pressures?

Salil Parekh:

So first thanks, thanks Rachel for doing this and for what you've all organized across the week, phenomenal. I think in fact the ROI question, what we just heard about how NASDAQ are using AI, the investment and I'm guessing, not knowing the specifics, is significant in terms of cost that they've put in to get that sort of an outcome. That was a bit of a forward view that the CEO, the board, the leadership at NASDAQ took. And the sense I'm getting is many of our clients, most people at large enterprises are doing that. The benefit of AI for people who are using it, there's huge productivity, but there's huge time to market, there's huge benefit in ability to scale. The sort of thing with 23-5, would have been less possible without some of the AI toolkit that they put in? And we're seeing that across the board. Consumer products company, we see them using AI agents to increase growth for their product sales because they are much more customized for what the consumers are buying by using that. It's basically as simple as using that AI from how I use that app through my social media usage and bringing more product to me directly, increasing their sales. And so the ROI question, there are traditional measures which will work extremely well at the end. Everyone who's investing massive CapEx has to generate some free cash flow. So that's a given but there are other ROI measures where for example the attractiveness of a large IPO to go to NASDAQ versus someone else because of these sorts of tools that they have built on AI will be a huge play and that's a growth driver. There are measures in terms of sensitivity to data, the security that becomes a huge growth driver and then there are measures in terms of fine-tuning, the flexibility that you can see with AI systems. So we see all of those sorts of ROI measures which are different from financial measures but the financial measures are not going away. Those absolutely there, these are enhancing some of that actively.

Rachel Oppenheim:

There's a lot of anxiety, a lot of dialogue. We're here in Washington, kind of at this great cross-section of policy and business leaders, but a lot of talk amongst that group about what AI represents for jobs. You employ, as I mentioned at the top, hundreds of thousands of people across the world. I wanted to just with that really get your perspective on what this means for jobs globally across industries and also how you're thinking about whether it's reskilling or upskilling or recalibrating within your own organization. I think there's probably a lot of really kind of dystopian views, maybe a lot of utopian views. Help us navigate through that. So there maybe start off with some data from within Infosys.

Salil Parekh:

We just finished our financial year in March. This last year we recruited 20,000 college graduates. And we have already in our plan for this coming financial year to recruit another 20,000 college graduates. And we have a base of 300,000 employees worldwide. 30,000 of which are in the US, and we're recruiting college graduates in the US as everywhere else. So what it tells us is what we are seeing with AI is new opportunities for work. So we today have our teams working with the foundation model companies for different AI applications to build agents, to develop software, for customer service, other areas. As we do some of the work in software, which is a big application with AI today, we see that people who don't have, let's say, software development background for 5, 10, 15 years, but have a concept of what they want to build. Let's say they want to build something more secure, more variable, more consumer oriented, they are able to use these foundation models to develop the code. And then with some help from engineering teams, test it and then make that ready to use in a public environment. So what it's doing in reality is expanding who can build software. And with that, we see people who have different ideas about what they want to build. Because in many of these cases, people from the business environment understand what they want to build. And traditionally, they would then tell the engineers what to build. Now, they can work with a foundation model and build it. And so you have an explosion of the types of things that can be done. To give you another data point, I was with the product development leader for one of the foundation models. And his view was, the amount of software we have is going to be 100x in the next several years. This is in the range of five to ten years. Not 5x, not 10x, 100x. So even if the productivity of. It's hard to imagine what that would look like. Everything we're doing is now software. I mean, anything you can imagine is getting converted into people building some app for it. So even if the productivity because of a foundation model is 10x, you still need 10 times more people. So that's the view that people in the business are working with. If you go with the static view, which is things will be as they are, then there are constraints. But at least my view, which we see in a little bit of data in the first year, is that there's going to be more of this growth and therefore more people. Now, the second part to your question was on skilling. Our approach has been to reskill everyone. We've not done retrenchment. At least in our business and that takes time because reskilling needs more work. But we think with the reskilling, with the foundation model, with other tools and many other tools, we think we'll get more impact going forward.

Rachel Oppenheim:

Another question for you just from your vantage point working with so many incredibly large companies who I know are all really racing to adopt this technology, but find that the data infrastructure, the data strategy that they have to actually kind of provide as inputs into these tools are, I think woefully insufficient. What are you seeing just that separates the companies who are really getting step one right, which it seems is a prerequisite to getting the most out of these capabilities?

Salil Parekh:

Absolutely. I think what we see for deploying AI at scale, there are two sort of components which are foundations. One is cloud and one is data. And data becomes really important. Because most large enterprises want to rightly protect their data. So they are not using the foundation models as we would use as a consumer, which is much more sharing our data and not having lot of control when we use a ChatGPT or Perplexity, everything what we do is essentially public. With a large company, they're using foundation models with their data completely in their own ecosystem. Nothing is public at all. Having said that, their data infrastructure is not up to speed because their data infrastructure typically is based on database technology but does not capture what we call unstructured data. So everything from their emails to their Slack talk to any consumer sentiment on social media, nothing is captured. And so one of the first things we end up doing is making sure there's an architecture for that data so you can have complete access knowledge and then making that completely secure because their public sentiment actually can be captured by anyone, because it's all on social media anyway. And how do you then make sure they make the right decisions with that data and then use it into the foundation? Because they use that data then to train the foundation model. Which then becomes relevant for that specific company. So huge, huge play on data and in fact, as a consequence we're seeing that data work expanding more. That's a positive benefit for Infosys.

Rachel Oppenheim:

So just a final question so many organizations, are really talking a big game about responsibility and governance and safety. How do you think companies can kind of really approach it as more than a box-checking exercise and to take it seriously? So we'd love to hear your thoughts as our final question.

Salil Parekh:

Absolutely. I think there we have seen already that one can build AI models, agents, which if not built responsibly can have tremendous bias in them. There are global approaches on how it's to be done without letting the bias creep into it. But there's bias in even how humans build software or humans develop systems. So in fact, to remove it from AI is even more difficult because it's not like we are removing it from how humans build systems or software. However, it is possible. And so we have a responsible AI method of checking that and then rolling out anything that we build. Almost every client we work with is aligned to that. So there's not any big discussion. The only thing we have to keep in mind many times is there's a lot of demand for speed and in that sort of getting things done, sometimes people may overlook the responsible AI framework. For the most part, we see that's being adhered to. And it's a critical element as we scale up. And you heard from the foundation model companies that they're using that as well as they build the models.

Rachel Oppenheim:

Salil, thank you so much for your time. It was a privilege. Thank you. Thanks very much.