Deploying AI – Techstrong AI Podcast EP23

Hello, and welcome to the Techstrong AI Podcast. I'm Amanda Ani, and with me today I am excited to have Krishna Mohan. He is the deputy head of TCS AI Cloud Business Unit.

How are you doing today? Doing great, Amanda. Thank you for having me.

Looking forward, uh, with the interactive session on AI and its impact. Yes. Happy to have you on our show.

And, and with that, we are going to be talking about a survey recently conducted by Tata Consultancy Services of 1300 CEOs and other senior executives across 12 industries. Um, and it, it shows that eight out of 10 86% of senior business leaders have already deployed AI to enhance existing revenue streams or create new ones. Um, but I'd like you to share a little bit more details about this survey and the impact that it has on business leaders.

Yeah, sure. Amanda, uh, so recently TCS has actually released the TCSA for business study, uh, what we titled as, uh, some potential to Performance by design. Uh, as you rightly capture, uh, more than 1300, uh, you know, execs, CEOs, and the p and l owners, uh, for large customers across the 12 industries, 24 countries.

We having, you know, uh, did this survey, and I think one, one thing I wanna call out before going into the details of the survey is also the timing of the survey. Uh, we all know, uh, AI and especially with the generative ai, uh, really caught the imagination of the whole world. Uh, but we didn't want to do the survey right front, uh, you know, being a consulting, uh, and services provider, we said, let's actually wait, uh, until the hype cycle slightly comes down a bit and capture the experiential, uh, views of our, uh, you know, senior execs of the industries that we sell.

Uh, I think that that, uh, kind of, uh, provided a balanced view, if you will. Uh, there is a lot of excitement, uh, in the long term potential of United AI and its impact, as you rightly said, eight, eight out of 10 felt. The long term view is to really improve business as business models, business KPIs and business outcomes.

That's the real power of ai. But also it's balanced by a lot of adoption challenges, if you will. Uh, while a lot of people have started, but they want to look at how to scale it, uh, there are also need, uh, only 17% I think of felt there is enough KPIs, but a lot of the people felt there is not enough KPAs to measure the outcomes through AI backed yet.

Um, and then almost some 50 people, percent of people are so felt that there has to be a global, uh, you know, regulatory framework, uh, to really take out any biases and, and, you know, and for whatever is required to ground. Uh, so, but I think it's a very experiential view, and it kind of resonates not only with these 1300, uh, plus customers, uh, even the folks who have not participated very well. So it's very well taken that way.

I think that was very wise to wait until the hype died down and get a, a better, more practical experienced view from business leaders. And this is an issue that many bus business leaders state is the return on investment and really understanding if it's doing its job correctly when they harness this AI technology. So what are some things that you're hearing from companies and business leaders as far as how do they solve this issue?

Do you have any tips or advice? Yeah, again, let me top from the experience. Very, very good question.

Uh, ROI is definitely a parameter, but not only the parameter, right? So because we are still in exploration phase, or I probably, uh, at the f end of the exploration phase, I would say, and we are seeing slowly, uh, you know, it's, it's getting into production, uh, because the, the customers that we saw, which are Fortune 2000, uh, you know, for example, one of the Wall Street customer I was talking, and his problem is a financial services customer basically said, you know what? I have 4,000 use cases.

So we formed an generic committee and let each of our lines of businesses given them the freedom to go ahead and explore. Uh, and, and we didn't want to definitely control from central, uh, you know, committee or a central, uh, unit, but we asked each LOB to explore. Now we have a problem of plenty.

We have 4,000 use cases. Now, I all I can probably take to production is, you know, maybe 5% of them are 10% of the use cases. So that itself is 40.

Uh, right? If is, uh, if it's even 1%, I think it's, uh, you know, uh, very less, but a 10% is 400, 5% is 200. Uh, can I actually scale it?

Uh, so the approach that we have seen few people and few organizations take an ROI is let me not look at each use case and each implementation for an LOB and ask for what the ROI can I, as in line of business, try out 10 use cases shortlist, and at the set of these 10, um, can I create a, a, you know, value, business value? Uh, so I think that freedom, again, this is not the success model as such. This is one of the model that I could immediately recall.

Uh, but the good news is though the entire industry is trying to, uh, bring down the cost of generative ai, uh, you know, you have seen Google, uh, Gemini Pro increase the maximum token size, therefore you can actually, uh, kind of use a larger set of data, uh, for LLMs, uh, either to train or to actually, uh, have them act on it. Uh, of course, NVIDIA and others continues to, uh, drive down the cost of GPU. So there is a lot that is actually happening to bring down the cost.

Uh, but an ROI perspective, uh, I, I believe the customers are moving from what I call, is use case to value case. Uh, they really wanna see what the value it delivers. Uh, savings is definitely a portion of it, but as the survey calls out the long-term value, they actually see the business benefits.

So as long as you're able to articulate it, uh, and I don't think you'll have right answers immediately, but take a set of problems and look at it, uh, and go after that and see what it actually can deliver. So that, I think that's what we're actually seeing it. Uh, Amanda, Yeah, that's a good example.

I think that's a good idea to reduce the amount of use cases, narrow it down, and then test those and, and really have a good idea of the outcome you want to achieve when you implement this AI technology. Just not implementing it for the sake of doing so. That's right.

That's right. So, do you think that more companies should have, uh, a department dedicated to AI technology? Is that something that should be a separate department?

What are your thoughts? See, there are two thoughts in it. Uh, one, uh, because it is still an evolving technology, um, and Gene definitely democratize the AI adoption.

Um, so what Gene did is CA used to be kind of a, you know, privileged, a few can actually play because it needed, uh, you know, you had scientists, you, you need somebody who can understand the machine learning and a much more deeper, uh, with gene AI is kind of made the adoption of AI as well increase not only generative ai. So overall AI actually adoption has actually increased. So there is one thought that, you know, why to, why should I control how the deployments to be done?

Let each of the businesses, uh, and of course the CIO comes into play. CTO comes into play, uh, more from designing the framework. A, this is, uh, probably the LLMs that we would use.

Um, right, e every day a new LLM is coming out, but at least for a, you know, for a retailer, it say, these are the 10 LLMs that we would use. Uh, and you have all the usual suspects, open ai, uh, anthropic, Gemini, and Google Gemini and, and few others in the open source. Uh, so 10 of them, these are the ones which we, we believe that should be used.

This is how it'll actually integrate with the, uh, other applications. These are the guide rail, at minimum, we should have. That's where I see the role of centralized, uh, units.

But the experimentation itself, uh, you know, they are letting the, uh, regular, uh, LOB of business unit try out. So if at all, if there is a centralized currently, um, it is more about the committees. You have committees who will decide what the risk.

Very rare actually for a technology, you have a committee which includes, uh, you know, the legal person, uh, lead into the committee because, uh, there is a, uh, still a, a kind of risk element to it. There, there, just wanna be careful. Uh, so that's how I actually see, uh, not very much consolidation, uh, consolidation and centralize the control more on frameworks and architecture.

Uh, but the experimentation and implementation is left to multiple groups. You mentioned the legal aspect, and there's still much to be done as far as laws and regulations in regard to ai. What are you hearing from business leaders as far as they're concerned about making sure that they follow current guidelines When it comes to ai?

I think the, uh, while definitely guide, uh, you know, um, the, the overall security, the legal and the IP rights, mainly, there's a fear that what happens to my IP can actually secure it. Uh, am I training the LLM on my ip? Probably is the biggest thing, uh, right where everybody has it.

Uh, but I should say that guardrails, that actually, uh, from a regulatory perspective that has been implemented, uh, right now industry broadly is, is okay with it. Um, you know, you saw Europe recently rolling out, um, U US is significantly, uh, you know, making progress on it. There are a lot of Senate committees that are actually being formed.

Uh, we are also actively contributing to those, uh, you know, regulatory things. So, uh, broadly I would actually say the entire, you know, market, the OEMs, the governments, the industry, the academia, all are coming together to define it. Uh, and whatever is right now there, um, it's, it's decent.

I would actually say, uh, but bit more, I think, again, it calls out, um, the, the AI for Business Study, uh, is divided in a way that 50% say that no government should put a control. Other 50% say no, it's actually the industry that should be, uh, controlling it. Arriving at the GAR RA is not necessarily government, but I think the answer is somewhere, it's all combined.

And that's the effort that we actually see. And there are, for example, Microsoft has implemented in open ai. I mean, you can choose, uh, not to have your data go to a, uh, an LLM, um, right?

So those controls already started coming up. Uh, few of the LLM providers are saying to an extent, uh, I will kind of underwrite you that this will not go, um, to the public. Uh, and, and we are seeing that adoption, large companies, whoever wants to build their own LLMs, they are actually implementing, uh, these LLMs in their own data centers with their, their own data.

They're ensuring nothing goes out to the public, uh, you know, outside at all to the internet or any, any, uh, not their data is not used to train the LLMs, if you will. So it's an evolving space in the right direction, everybody coming together. That's, that's how I see it.

Absolutely. Now, are you still hearing from business leaders that there's a skills gap when it comes to the AI technology? And how would you best address this skills gap?

Would it be through some sort of education or training within companies? Uh, how can they find the talent that they need? Uh, I think skills gap is, uh, is real.

Uh, so not many people, probably 5% of the overall team ever got trained or competent on AI machine learning. So far now with generative ai, and we are actually looking at these implementations in reimagining the IT services, reimagining the bus business processors like finance, hr, uh, right. And on the business side itself, the, for example, claims reimagination is happening, um, you know, regulatory, that development actually is actually happening.

Regulatory filing is happening. So given that gene is, uh, all pervasive, so now the need to train everybody on what, what it is, because the role of the person who is doing the work, uh, is got a change. So it not only that you need to get trained on these LLMs and the models, but also about how to implement it, how it changes, the change management, the processes, everything actually is gonna change.

So that is definitely currently, uh, you know, we, we definitely see that as a problem. Uh, and it is a challenge, and I mean, companies like us, TCS, so we have ourselves, we train half of our 600,000 on Gen AI ready, uh, so we are there to help. Uh, but again, uh, it's not that one time you train, these things are really evolving.

So there's a continuum of training, uh, is definitely there. And I think it'll take, uh, you know, some time for everybody to be comfortable with it. Uh, this is just a, uh, maybe I'm repeating myself.

It's an evolving space. Uh, therefore, uh, we need to look out for another 18 months, uh, of how this thing evolves and continuously train people. Yes, I think everyone would definitely agree.

This technology has advanced rapidly since coming on the market a couple years ago. So if there was a key takeaway that you could leave with our audience today, what would that be? I think it's a great space to experiment.

Um, you know, and there is a very clear, the companies that have significantly progressed on cloud are also the companies that are actually leveraging AI much more and efficient and in much more advanced way. Uh, what I call is the, you know, progressive customers, uh, that I actually see. Um, and, uh, you know, there are models that are platforms that are coming out where while there is plethora of LLMs, uh, but you know, TCS is coming back, recent, uh, recently released the TCS Wisdom Next as a platform, which actually helps the people to choose what is the right model for the specific use case and what are the right costs, right?

With this, you know, and all built in already. Uh, so there is an evolving space. All the elements of it is actually being, uh, handled well.

But if I had to take away one thing, uh, as a customer, as an industry, is moving from use case to value case and parts to production is the most important parts to production, to value is the most important thing that we see, uh, customers, uh, going through in this exciting times ahead. Of course. Yes, it is.

Well, thank you so much for coming on our show and giving your insights. My pleasure, Amanda. Thank you very much.

Uh, have a great, uh, July 4th holiday To you. You too. And thank you to our audience for staying tuned this week, and we look forward to doing this again.

If you missed last week's episode, it's there. Go back and check it out and leave us some feedback. What are you interested in hearing?

Have a great day.