Integrating Data for Digital Transformation: A Conversation with Andy Logani, EVP and Chief Digital Officer at EXL | Podcast Ep. 15

We used to be a business process management company. Then we evolved and became analytics and a business process company. Then we embedded digital and integrated that into the workflow.

Then subsequently, we became a data-led company. And most recently, we are a data and AI-led company. So, this just shows you, our evolution.

So, over these 24 years, every time we’ve evolved. But the interesting thing also is every pivot has been faster than the previous pivot. So, if it took us five years, the next one took us four.

The recent one took us three. So, every pivot is faster. But thankfully, we’ve been able to add more value to our clients.

And we’ve been able to also increase our growth rate as a company and double-charge our growth because we always evolve with the trends. My own personal journey has been quite fascinating with the company. I was in the business.

Then I spent a lot of time with analytics and data. And then most recently, a couple of years back, I took on as a Chief Digital Officer. The big portion of my mandate is to take data, take AI and the newest technologies, combine with our industry expertise and deliver value to the customer.

Host: That’s very inspiring. So, speaking of data, what role does data play in today’s technologically progressive commercial setting, especially in informed decision-making?

Andy: That’s a great question. And thankfully, with the recent advent of generative AI, this thing has come to the forefront of the board and the executive conversations, which people knew the importance, but it wasn’t the top priority for them. Specifically, to your question, if you do not solve for data, you actually will not meet the promise of your AI.

You will not meet the promise of your transformation. And you can buy the most expensive technology. You will not deliver the benefits to your customer or the problems you’re trying to solve.

So, the most important thing that you solve for is actually for data. Historically, what was happening is, people will just put a lot of emphasis on buying the latest and greatest technologies. And we saw 60% to 70% transformation efforts failing is because there was no data.

I’ll give you a very quick example. This applies to mostly all industries that we work in, whether it’s insurance or healthcare or banking, to name a few. One of the most important things is to know your customer well or have a better Customer 360 profile, so that you can personalize the experience, not only how you deal with them, the choice of products you offer to them, the choice of experiences you offer for them.

And a lot of companies were not able to even put a good Customer 360 view in the first place. And what was lacking is actually the foundational data. One of the recent advancements that makes this even far more important than it ever was, in a lot of industries, people may still have solved for structured data, which is you can get the data organized and you can get it structured, you can get the information flowing.

But a large pool of intelligence was actually in your unstructured data. And that unstructured data has now gotten unlocked, especially with the advent of generative AI. And that has become key to decision making.

So, the power of combining both your structured and unstructured data, feeding that data in a privacy-preserved way, completely high quality, making sure it’s available when you need it, bring that insight. And when that insight gets orchestrated to the workflow, that’s where the magic happens. So, for us, data plays a very, very foundational role in making sure that you meet your AI or your digital transformation promise.

Host: So, as we all know, digital transformation is a very important trend for business nowadays. So, in your perspective, what key challenges generally arise during such shifts?

Andy: Sure. Now, again, another excellent question. See, one thing I will tell you, which is, actually, it’s quite bizarre, but it’s so true.

The number one pitfall even today is that people just go, sometimes do things for the sake of testing the technology versus really asking the question, what value it will bring and what problem will it solve? Actually, you’ve got to start from what customer journey it’s addressing, what promise it’s going to fulfil, and actually what value it will deliver. And we see often that step gets missed.

It’s like a hammer chasing the nail. So that’s like one big challenge. Secondly, look with any technology.

By the way, that’s true to AI as well. Everybody gets excited about artificial intelligence, but people are learning when you scale it, it takes time, it’s not easy, and there are its own challenges and impediments as you deploy. And one of the other things that people trivialize or sometimes underestimate is the effort that you need to execute the change.

So, change management from going to point A to point B, we still see as a challenge, and we still see a lot of, unfortunately, our clients end up trivializing it and that doesn’t help them. The third thing I will say is don’t take technology and take it to, okay, I will use this technology and take this forward to an outcome. It’s actually what journey and what design am I creating, what reimagination I’m creating, and then work backwards.

So, technology plays the role to reimagine in how you’re going to conduct the business versus here is the technology and I want you to embrace it and adopt it in the manner I want you to. So, I would say it’s still a proper design and understanding the journeys becomes very important. I think the last thing I will say is not anything in particular will solve everything.

Now, people suddenly hear about generative AI and they think, oh my God, generative AI has come and it will solve all my problems. Prior to that, there was machine learning, people thought that will solve all my problem. Prior to that, there was predictive analytics and people thought that will solve their problem.

Actually, it’s the combination, it’s the orchestration and it’s the stitching. You’ve got to look at what thing you’re solving for. You stitch the right technology; you bring the right data and you bring it to the right context.

That marriage is what creates the magic. And I still see a lot of gaps because this requires a harmony and like a symphony, like you do an orchestra, right? You have to bring it together and stitching this together is an art.

One of the big roles that we play is actually stitching this together for our clients at the end of the day so that they can get the value or the business benefit that they’re trying to deliver. I hope that sort of gives you a sense of some of the challenges that we still see our clients encounter in this journey.

Host: Yeah, that’s really a great perspective. Moving on, how have recent innovations contributed to the advancements of data analytics? Can you talk about a few technologies that have remarkably influenced the impact of data analytics?

Andy: Sure. Actually, I’ll break this into two or three buckets. See, first and foremost, just on the data side, there are so many technologies that have now come up, you know, like you have the data mesh, the data fabric, ability to generate the synthetic data, data lineage, data observability, creating a lake house infrastructure.

You know, so there are now platforms that you can work with. So like we partner with Databricks as an example. We also partner with other companies, you know, like for example, Microsoft Fabric.

And we take these technologies and great technology platforms and that gives you a great foundation. This is enabling, you know, good data, clean data available in the right infrastructure. And most importantly, for you to scale it when you need it, right?

On the AI gen AI side, you know, amazing advancements already. You know, you’re hearing about generative AI, for example, its ability to harness unstructured data. That is a fundamental game changer on AI.

We are now actually working with a client on agentic AI or agentic workflows. Just to give you an example, with this banking customer, we’re going to actually take a third of their calls, identify all the intents, and then actually action using agentic workflows, completely eliminating the need for human involvement. And we are using obviously autonomous agents.

We are using large language models. We are using appropriate guardrails and we are making it available to an end-to-end stitched platform, right? So that’s kind of another amazing advancement on the AI front, the agentic workflows.

Obviously, most recently, multimodal, which is earlier, the whole advancement was around text and you were leveraging text only to retrieve intelligence, right? And even some of the early versions of the large language models were very focused on text. The transformer models earlier versions.

Now it’s multimodal. It’s image, it’s video, it’s text. And you can actually take these technologies and fundamentally change the experience, right?

And look, slowly but surely, I know there’s a lot of talk about the AGI, whether it’s going to get there and completely sort of replace the human. I personally think that we have some ways to go. We fundamentally believe that combining data with the appropriate AI advancements with industry context and stitching that together is where the magic happens.

But look, several of these technologies have created a great way of doing things. One other example I’ll give you, we are in this business of assessing claims damages. So first and foremost, the pictures of the roof damage, we replace the need for engineers to go there.

It’s actually captured through drones, which even local teenagers can get the pictures for us. You take these images through computer vision, and this is like classic machine learning, computer vision types of some of the NLP programs. We are assessing what’s the extent of the damage of the claim.

We’re also assessing whether it’s a hail damage or not a hail damage. So the programs are advanced. Now you’re taking that, you’re assessing what would it take to sort of settle that claim.

And then you can completely automate the experience, orchestrated the workflow and adjudicate the claim. So what used to take such a long time has got shortened completely. You’re capturing this data, and there is enormous compute now to process that data.

You’re using appropriate techniques. And I just gave you this example specifically that, look, not all problems you need generative AI, even some other technologies and classic AI machine learning models work really well. The whole focus is, again, I will sound like a broken record here.

What problem are you solving? And you work backward from there and stitch these things together to solve the problem for the customer.

Host: All right. So, Andy, you and your team recently disclosed EXL Insurance LLM. So, could you provide an overview of this facility and its objectives?

Andy: Yes, yes. See, look, there’s like two schools of thought, right? One school of thought is pre-trained LLMs are advancing so rapidly.

Why would you fine tune and create smaller versions of large language models or call it small models or there are various purpose-built models. There’ll be various names associated with it, right? In this particular case, we actually first worked with a pre-trained model and we realized two or three things which we felt that we can overcome by fine tuning a purpose-fit model, which is for the insurance industry.

But I’ll put this in context. The reason we were able to do this is because for the last 20 years, we’ve been processing these medical records on bodily injury claims, commercial liability claims, personal auto claims. We actually have Q&A tagging.

So, we know how this response works for specific questions. We have procedure manuals and modules in how these transactions are processed. So, we had such rich domain history and knowledge of these businesses.

So, what we did is we took a pre-trained model. We completely fine-tuned by bringing our own data, which is this data of rich data that we have over the years on Q&A pairing from the medical records, our knowledge corpus, our understanding of processing these claims. What we found is we overperformed the pre-trained model on accuracy by 30 to 40%, depending on the model.

We also started to create much faster speed because with the pre-trained model, we had to put such massive prompts. It was taking too much time. And then the response was also taking a lot of time.

So, we were able to cut that short. We are also able to bring the cost down significantly. So, you know, when you work with these things, you’re always thinking what latency I want to have, what accuracy I want to deliver, at what cost it will be done.

And obviously we’re going to all do this in a very, very privacy-preserved way to protect our customer’s data, to protect the PII, to make sure everything that we do is done in a responsible manner. So, we saw outstanding results and therefore we applied and we are seeing great results. I’m happy to share with you.

Two of our clients are going to go live with this new insurance LLM. We are creating more variations for it in underwriting. But again, we are very agnostic.

Our idea is not to just create LLM for everything. Wherever we see there is value and it brings value to the customer, we will absolutely do it. So that’s kind of the background to the insurance LLM and what we’ve just launched in the market.

Host: That sounds very exciting. Moving ahead, congratulations, as EXL has completed 25 glorious years since its establishment. So how do you plan to progress further in the upcoming years?

Andy: Yeah, look, it’s amazing that we’ve had this 25-year journey and I’ve been with the company 23 years. I never thought that I’d last that long and how many different things I’ve been able to do. It’s just an amazing company to work for and I’m still enjoying it.

We are refreshed every day taking on these new challenges. The other amazing thing about the company, which I shared with you early, is we’ve kept making pivots because if you stay static, then you write your own destiny and we don’t want to become a dinosaur, right? So, one essential fabric of our culture is constant reinvention, constant upping the continuum, keep going up and keep exploring newer things and better things for our customers.

But we’ll always stray true to two, three things. Where is the value? What makes the most sense for the customer?

And where do we play and what role do we play in that journey? So, for example, we have made a strategic choice to be a data and AI-led company. We are allocating a lot of capital.

As you see, we’ve made recent acquisition in acquiring these data companies because we want to be known in the business, be able to help our clients. We are also making significant advancements in AI. So, as we speak, I’ll just give you a couple of examples.

We have launched in our partnership with NVIDIA three solutions. We are working with other hyperscalers like AWS and Google as well as Microsoft and we will be launching our solutions on AWS and on Azure Stack. We’ve already made several of our solutions.

And when I say solution, it’s the data, industry knowledge, AI coming together and available to customers on a pay-per-use basis or on a SaaS basis. So, we have launched several solutions already and we’ll continue to be deep into this ecosystem and continue to innovate, right? Because for our business, we just have to continue to innovate.

So, we’ll keep making purposeful bets. We’ll keep making purposeful reinventions, but at the end of the day, we’ll always add value to our customers and we’ll keep innovating and keep upping the continuum.

Host: So, Andy, according to you, why is it crucial to adopt risk analytics strategies in digital operations? And how do they help us stay compliant?

Andy: Yeah, look, especially in this new world, the dimension of risk and with every evolving technology, if you don’t do it a responsible way, the risk goes up also significantly. And one of the things that we always do is we think of a risk from a complete spectrum in terms of how are we stitching these pipelines? How are we bringing data?

How are we taking AI to data or bringing data to AI? How we do it in a responsible way, right? And we’ve actually put several solutions.

I’ll give you a quick example. Recently, when we were working on this insurance LLM, we wanted to make sure that the data is completely PII free. We wanted to make sure that there is a complete anonymization of data.

And what we found that the utilities that are available in the market are only taking you to 70% anonymization. We created our own ground truth model, merged it with what was available on these market offerings. And we took the anonymization as high as 99.5% because we care for the privacy and for the customers, right?

Similarly, we have also put lines of defense in our digital operations. All the transactions that we do, we are now putting AI-based mechanisms of auditing those transactions or those operations so that AI is helping us detect any patterns, any trends. If there are any anomalies, we are incorporating smart signals in which we get an early warning detection that, hey, if this happens, you could have an exposure.

So, there are things that we are doing in us, like things like smart audit, which we embed in our digital operations. Then even take it one level up, which is actually a level third line of defense. Even in our audit posture, we are using AI-generated AI to even reimagine how audits get done so that you are looking at risk in all three spectrums of lines of defense, level one, level two, and level three.

And particularly on risk analytics, not just for EXL in our own operations, we actually have a complete market offering for risk officers in banks, insurance industries, healthcare, all the segments that we work, where we help our clients understand the risk quotient and use data and analytics, help them mitigate and manage the risk better. So, it’s a market offering as well as what we do internally for our own operations, digital operations.

Host: So lastly, considering the potential of artificial intelligence, what future opportunities and difficulties do you think it will unlock for business across industries?

Andy: Yeah. Look, one of the most exciting things about this AI-Gen-AI evolution is the pace at which it’s moving. The problem that exists today gets eliminated tomorrow, right?

A lot of people said LLMs can’t solve for math. A lot of people said LLMs can’t do reasoning. Look at where we were and look at where we are.

And by the way, we started working with Transformer, GPT, 2.0 models like four years back. So, I’ve seen this evolution so quickly and last two years particularly have been amazing. So, the potential of what this brings to us is absolutely mind-boggling.

But if you don’t do this in a responsible way, like we discussed earlier and all the mechanisms to do this in a responsible way, then you can also bad actors and this falling in the hands of bad actors can create a lot of problems for us also. So, we are very mindful of minimal data, minimal technology principles first. So, use what you need.

Make sure that everything that you’re stitching right from the data pipelines to AI pipelines is done in the most privacy-preserved way. We put additional guardrails to make sure that even if you put the perfect pipelines, there’s always ways and means where things can happen. So, we put guardrails.

And then last but not the least, a human in the loop. We are still not at a point that you’re going to leave this completely on its own. In some places, it’ll go there gradually for some journeys, but augmented intelligence and the role of a human is still very important to oversee and have that human in the loop layer, right?

So, we consider this as an amazing opportunity, but absolutely need to be done in a very responsible way.

Host: Well, thank you so much, Andy, for sharing your valuable insights and experiences on technology and digital transformation with us. It was a real pleasure to host you today.

Andy: Thank you very much for the opportunity. I really enjoyed talking to you. And thank you again.

Host: Thank you, everyone, for joining us today. I am your host, Sayali, signing off. See you soon in the next episode of ExtraMile by HiTechNectar with our next extraordinary leader on board sharing their thoughts and knowledge.

So please stay tuned.