Gautam Singh

The number of opportunities and use cases that can be tackled have increased manifold.

Raffaelle Breaks

…WNS’ GEN AI Intelligent Document Processing Platform.

Divakaran Ullampuzha Mana

A successful agentic AI system is the data.

Gautam Singh

What are the few things that organizations are getting wrong?

Divakaran Ullampuzha Mana

…which talk about the success rate implementing agentic AI.

Raffaelle Breaks

…provides a lot of data accuracy.

Divakaran Ullampuzha Mana

What business problem they're trying to solve?

Gautam Singh

Hello, everyone and welcome to this fireside chat. I'm very pleased to welcome my fellow participants, Raff and Diva and myself. We'll introduce ourselves shortly.

But the topic for discussion today is driving business outcomes with agentic AI-powered SKENSE. We will also shortly introduce what SKENSE is. But without further ado, let me start with introductions.

Raff, do you want to go first?

Raffaelle Breaks

Hi everyone, I'm Raffaelle Breaks. I'm the Chief Product and Design Officer here at WNS.

Gautam Singh

Diva?

Divakaran Ullampuzha Mana

Hi, my name is Divakaran Ullampuzha Mana. Diva in short.

I lead the solution architect team for AWS India for all of the ITES customers.

Gautam Singh

Thanks, Diva and Raff.

To introduce myself as well, my name is Gautam Singh.

I'm the Global Head for Data Analytics and AI at WNS, and I'm based here in the UK. Raff, let's just cover what SKENSE is for everybody's benefit, and then we can go into our conversation.

Raffaelle Breaks

Great, thanks Gautam Singh.

At the heart of it, SKENSE is WNS’ intelligent document processing platform. So, we have a fantastic platform that meets the needs across a wide range of organizations and industries to systematically manage large volumes of unstructured or semi-structured data.

It's essentially solving for a lot of the manual data entry that leads to inconsistent data quality. If we think about how organizations on their AI transformation journey need to take a lot of this unstructured or semi-structured data and put it in a presentable format into either systems of record or into workflows. So, it's really enabling them with an AI powered solution, use of AI, both NLP and ML to automatically extract and validate data.

It provides a lot of data accuracy, obviously reduce costs and efficiency, which is everyone wants. But it also enables faster decision-making within the organization because it provides the insights and the data points to enable decision-making within the organization. And it also has inbuilt workflows to ensure traceability, auditability and AI responsibility.

Gautam Singh

Thank you, Raff. Thank you for summarizing that at the very start. I'm sure we'll find out more during this conversation as to how agentic AI has been incorporated into SKENSE, and what role AWS is playing on that front as well.

But let me start with a high-level question. And let me ask to both of you, with all the increasing interest around agentic AI, in your view, what are the few things that organizations are getting wrong? I think this will set a good backdrop to our conversation.

Raffaelle Breaks

I think from a business perspective, and everyone's heard about the different stats around the number of POCs that everyone's been doing to really leverage AI within their organizations.

But really at the foundation of those AI transformations, for most organizations it is going to be about how good is the data? How good is the underlying data that's powering their AI applications? And some of those organizations, what we're seeing is that there's not only data debt, but also tech debt and process debt. And so, to really take and make the most out of agentic AI, every organization needs to look at their data. They need to look at their systems, and they need to look at their processes.

And in addition, identify the right use cases that qualify for agentic solutions. There might be some use cases that can be solved with ML. There might be some, though, that require, and these, I think, is where agentic AI can really power more modern solutions.

Because in the past, there's been a lot of nuances, or there's been organizations that haven't been able to transform digitally some of their processes because they haven't unlocked their data. They haven't redesigned some of their processes. But with agentic AI, you can actually implement more nuanced use cases and solve harder problems.

Because now, you can use the latest and greatest models and train for those different opportunities, which allows you to create more complex solutions.

Gautam Singh

That's very interesting. And I'm sure we'll be of even further interest to our audience as we get into the depth of our conversation. Diva, do you want to bring in AWS as well?

Divakaran Ullampuzha Mana

Yeah, and as Raff mentioned, I think there are lots of numbers which talk about the success rate of implementing agentic AI solutions. And in my discussions with multiple customers in India, multiple customers outside India as well, there are typically two, three major patterns that evolve.

The first one is what I call solution-first syndrome. What it essentially means is that customers rush into building an agentic AI solution without really first defining what business problem they're trying to solve and what is the outcome they want to achieve.

So, they just go ahead into building an AI solution just because they have to. Either it's a mandate from the top or it is a requirement that they think the market needs. And what happens in the end is that it just lives as a POC and it does not go into production. It cannot scale or it cannot get the value that the organization requires.

The second problem that we typically see is something called evaluation blindness. So, while customers say, I want to implement or deploy an agentic AI solution. And when we go and ask them, how will you define or how will you know that the agent is doing its job well, there is total silence. A lot of times customers do not define, plan, design their evaluation systems and the metrics appropriately to really measure how an agentic AI system is working well.

Unlike a traditional software where you can measure maybe response time or error rates, with AI agents, you need to understand why they are making certain decisions, how they are planning these actions, and whether they are reflecting and learning appropriately. So that's evaluation blindness, which many people miss out in the early part of the process.

And the third problem that we typically see is the integration nightmare. As Raff said, the core of a successful agentic AI system is the data. And what we see is that many times while they have brilliant AI agents that they build, the data is trapped in silos. They are not able to access the data because it's either behind custom APIs or proprietary interfaces or legacy systems. And hence, the AI agent cannot do what it needs to do and cannot be as helpful as it can be if the data is all accessible to the agent as well.

So, these are broadly the three patterns that typically results in a lower rate of success as far as agentic AI implementation is concerned.

Raffaelle Breaks

Great. I mean, we've spoken about how and when to leverage.

But Gautam Singh, obviously, you've been working with many customers and clients. What are the business opportunities that agentic AI can deliver from your point of view?

Gautam Singh

I'm actually very excited by all this development and the pace at which things are changing. If I go back two or three years, we started with LLMs and chat GPT and all the noise around that. That was a game changer. And now we've got agentic AI, which is really a 2025 invention.

That is another leap forward in terms of the possibilities. So, from my perspective, I think the key outcome or key intervention because of agentic AI is the fact that we are no longer looking at process improvement as a continuous improvement, but really a process reimagination. And this is facilitated by this new technology coming together as agentic AI rather than just AI or Gen AI.

So, what does that translate into? A, I want to pick up on what Diva said, that we need to start with the right use cases. So, when I look at it from a process or workflow perspective, there are many more complex use cases that we can now put under an agentic AI consideration that we would have historically put forward as looking at it more from an automation-only perspective or an AI-only perspective.

So, the number of opportunities in use cases that can be tackled have increased manifold.

The second thing I would say is that when I look at how and what we do to those use cases, at the input end, at the data end, it's not only about harnessing data that sits in legacy systems, but there are also new sources of data that can be brought in to take advantage of what agentic AI systems can do, which is why I come back to reimagining the process rather than just tweaking it.

And new sources of data include video, audio, social media, actually even more unstructured data from sources that we wouldn't have normally considered, like satellite imaging, etc. And when you bring all these data sources together, apply an agentic AI proposition or solution design to it, the process that you're building is a reimagination rather than a re-engineering tweak.

Hence, the opportunities are quite endless, at least here at WNS. We have identified, and I'll talk a little bit more as this conversation progresses, but we've identified a number of different areas where we are adopting and adapting agentic AI solutions. Let me pause at that point, and we'll pick up more with specific examples as we talk.

Raffaelle Breaks

Yeah, I love the idea of process reimagining. Who wouldn't want to reimagine some of the processes that many organizations live with every day, right? And be able to think through and redesign those. It's a great opportunity.

Gautam Singh

Yeah, and since we're talking about process redesign, and how we would do it five years ago or ten years ago versus now, why do you think agentic AI and not just AI or just old-fashioned automation for some of the processes or use cases that are being considered?

Raffaelle Breaks

Yeah. So, having spent two decades digitizing a lot of processes at a large financial services company, what I think the difference now is, that was really, and that's why I really love this notion of reimagining processes. We were taking existing processes and automating them, putting business rules around them, presenting them in a digital journey, either through mobile or web. But what agentic AI really offers is that evolution from automation, business logic, to supporting redesigned processes that allow and require autonomy.

And, with both humans at the helm and human oversight, you can actually implement specific agents that are doing multi-step processes and much more complex processes, much more where there's things that require a lot more multi-step, but also some of the policies and procedures that companies have support because they want to drive revenue, and they've actually got behind the scenes, making certain business decisions. And agentic AI allows you to make that programmable, but also learning the continuous learning, evolution and iteration, and they can solve for that changing business context. Whereas when you were building like a web application, as an example, you know, it was a very linear flow.

It was like you start here, you add some logic and this is the outcome. And now the agentic AI and multi-step processes and that continuous learning and humans at the helm, it allows you to really reimagine, as you said, Gautam Singh, those processes in a way that best suits the organization and the business. And I think that's going to be really powerful and lead to a lot better outcomes for those organizations.

Gautam Singh

I think from our side, we've also developed useful frameworks to help things through, how we approach agentic AI solution to various use cases. And we've got a GAIN framework, we've got an EDGE framework, etc. And these are sort of consulting tools, if you like, so that we can work with our clients and walk them through the logic of first evaluating whether a use case is fit for agentic AI, but where it is, how do you approach it? How do you decide what agents are needed? What level of autonomy to give to them? How do we execute beyond that autonomy from a learning and memory and action from the output perspective as well?

Diva, maybe you want to come in from an AWS perspective, how AWS is taking care of some of these aspects?

Divakaran Ullampuzha Mana

Absolutely. So, I think very valid important points that we heard around process reimagination and looking at where to use AI and where to use agentic AI. From our perspective, when we go and meet customers, the mental model that we typically follow is as follows. One is we look at the difference between agentic AI and the traditional AI, if I can use that word, is to see the level of autonomy that is required and the level of continuous execution that is required. So traditionally, AI is reactive where you give a prompt and it gives you an answer, whereas agentic AI is proactive and autonomous.

So, we look at that angle, and then see for a particular step, for a particular set of tasks, that needs to be done or a process or a sub process that needs to be automated, does agentic AI play a role or does not play a role? Second, we look at the number of steps and the latency sensitivity of the requirement as well. So, if you're dealing with something that is latency sensitive, single step tasks, typically, we start with traditional AI, and most of the time, that might be sufficient. And if not, then there is a need to handle complex workflows, there is a need to adapt on an ongoing basis to changing conditions and operate continuously.

That's where agentic AI shines, and it's mostly apt for that requirement. From a human-AI perspective, we also consider complexity of the workflow. So, if it's a simple question answer scenario versus multi-step branching process, then agentic AI probably is a better fit.

Similarly, is there a need for continuous operation? Is it a one-time response and then go away? Or are there going to be ongoing autonomous task execution and continuous conversation that's happening?

Similarly, adaptation requirements between humans and AIs. Are they static responses that you can just code instead of using models and adding cost and memory and other stuff? Can you just code in the static workflow, I mean, flow of the conversation? Or is there a need to learn and evolve the behavior through conversations? And similarly, the complexity to integrate the various systems and the memory and the data sources, etc., also is a factor that we consider. So, these are typically what are some of the factors that we consider.

And on AWS, the advantage that we provide is that we provide a full spectrum of services, whether we're starting with Amazon Bedrock or SageMaker for traditional AI use cases, or move to Bedrock Agents or Agent Core for building customized agents that are fully autonomous. Or you can use Quick Suite, Kiro, Q Developer, etc., which are ready-made agentic solutions. And it allows us and the customers to meet at whatever point in the reference framework as needed.

Gautam Singh

Okay, excellent. Maybe we can jump into SKENSE now. And so, Raff, how does SKENSE play a role in helping organizations scale through such agentic AI or AI solutions?

Raffaelle Breaks

Yeah. What I love about SKENSE is its ability to scale to meet the needs of our customers and clients.

So, with one particular customer, I'll just give you an example before I jump into the technical aspects of it. But just to give you a sense of the scale, with one of our customers, we started with one business process and helped them move their unstructured, semi-structured data into a structured workflow that they could then make business decisions on. We started with one, and then as other parts of the organization saw the positive outcomes, results they were seeing, we actually have now scaled it to over 29 of their different processes.

And so, I think it's really impactful in terms of the amount of different documents it can support. So, just thinking through the different file types alone, so it can extract from PDFs, images, spreadsheets, slides, but it also can extract data and provide insights from videos. It supports over 150 different languages, and it can extract data from all the things from sort of massive tables to doctor's notes, which everyone knows are totally illegible.

And then it has, in terms of scalability, it's got an array of features and deployment options that are designed and built to be secure, accurate, and high-performing. And of course, we've architected it in a way that really, it sits very well on AWS. So, SKENSE can be hosted on AWS and leverages Amazon Bedrock, Amazon Textract, Amazon SQS, Amazon Simple Storage Services, so it's fully integrated.

So, as our customers and clients that are hosting on AWS, it fits in very nicely with their existing architecture. So, it can do the intelligent document ingestion, it can do the data extraction, it can provide intelligent workflow automation. And by combining those things, it provides the automation, the secure APIs, and delivers enterprise-grade solutions, which I think is really important because we work with many clients and customers, and it helps them empower them to look more extensively across their organization.

And as their business needs evolve, we've got a solution that scales with them, and it's built sustainably. Really importantly, it's built sustainably within AWS, and very high energy-efficient cloud platform, powered with renewable energy. I think that's really important as many of our customers and clients and ourselves are thinking about how do we make sure that we're creating solutions that tend to count possible impacts on our external environment as well.

Gautam Singh

Yeah, actually, it might be useful for me to bring a real case to life here. So, we have an insurance client where we are leveraging and deployed SKENSE-based agentic AI propositions. And this is a situation where in insurance companies, they’re below a certain triaging process, recoveries that they think are not worth fighting, they just basically pay them off.

So, claims that are not worth fighting for, they pay them off. What we've done with an existing insurance client is looked at their closed book of business, which is basically claims that they have paid off. And we have gone and said, okay, you know what, we can take these claims on a gain-share-based model.

In other words, we get paid nothing, but we'll take a cut of what we are able to salvage. We've applied AI models that are built on an agentic AI SKENSE-based platform. And those AI models are identifying from those closed set of claims opportunities, where frankly, they shouldn't have been closed, or the insurance company can give revenue back.

So, we've taken that on, and we've applied SKENSE, and the AI model that we have built that is both self-learning as well as built into SKENSE, and identify those recovery cases. We then put a human team to actually go and complete the workflow, which is to actually contact the insurance company or the counterparty insurance company, give them the evidence as to why this claim needs to be paid off by them and recover the cash into the bank. So, all of this is managed by WNS as an end-to-end workflow powered by the AI models that are identifying the claims cases that should be chased.

And that's further powered by an agentic AI solution within SKENSE, which is taking into account, taking, ingesting in data from the contractual side, from the images and history of the actual accidents or the claim itself back to the insurance company, looking at past behavior of the claimants, etc., etc., applying an agentic AI solution on this and powering the human claims agents to then go back to the counterparty insurance company and get that money back. It's been so powerful and so profitable, both for the client and for us, that the client has now incorporated this into their ongoing workflow. So, not just for closed cases, but for open cases.

So, the entire workflow is now being reimagined so that all claims go through this agentic AI-based workflow. And we've significantly improved both the recovery and the efficiency of that process. Okay, let's move on.

Raff, let's just jump into the benefits of leveraging the AWS stack for both agent, agentic AI and AI based solutions on SKENSE. I think you sort of covered it, but maybe we're just going into a bit more detail.

Raffaelle Breaks

Yeah, I think I did cover sort of some of it, I think just to reiterate and summarize.

So, in terms of the advantages, I think, SKENSE in itself is providing that automation and agentic AI-powered solution for document processing across many domains, including insurance, travel, healthcare. And, you know, we’re offering AI/ML power data extraction and understanding. But where we really have seen the benefits of AWS is that multi-agent collaboration across the AWS stack.

We're creating, scalability with other parts of that stack. And the integration and interoperability is going to be really impactful and important for our customers and clients. Obviously, the built-in security and compliance, and we've got rapid deployment, and the availability on AWS Marketplace, of course, also ensures for a smooth procurement process because who wouldn't want that?

We've all got enough things to deal with and managing sometimes tricky procurement processes being cut that time down and really bring this to market as soon as possible for our customers and clients. And we've seen that with being available on AWS Marketplace. So, it's a fantastic partnership with AWS.

Divakaran Ullampuzha Mana

Let me add a few more points on from what I've mentioned from my perspective, from AWS perspective as well. So, marketplace definitely is a great advantage, especially the reach to the larger market ecosystem that AWS has and the simplification, as Raff mentioned, simplification of the procurement process, the deployment process, etc..

But on top of that, a few other points, advantages that you'd probably have is one is the powerful set of comprehensive AI, agentic AI portfolio that we offer. Most of the offering, services are offered on a pay-as-you-go basis, which means you can start small and scale up and down as you need as your business requirements come up and go. Also, as I said earlier in answer to the previous question, we also offer not just the infrastructure components, but components across the stack.

And, you know, depending on whether a use case requires you to use something off the shelf, ready-made off the shelf, or you want to build something custom, or you just want complete control and just use the base infrastructure, the entry point can be any of those areas. The second biggest advantage, I think, is around security and governance foundation. And Raff, you touched upon security and compliance.

I just wanted to highlight that AWS is the first major cloud provider to have received ISO 42001 certification for responsible AI, which makes it, you know, which increases the trust and provides the foundation of trust that enterprises typically need for any autonomous system. And the other key point is around open standards commitment. So, our support to protocols like MCP or A2A and frameworks like Landgraf, Crew AI, etc., via Bedrock Agent Core ensures that SKENSE can evolve and integrate with future technologies without vendor lock-in.

The third point is around proven scalability, right? The combinations of services like Bedrock, Textract, S3, and Lambda provides virtually unlimited scale and enterprise reliability as well. The bottom line is that SKENSE on AWS delivers not just technical capabilities, but business transformation with security, compliance, operational excellence, and enterprise demand as well.

Gautam Singh

Thanks, Diva.

Actually, while we are on the topic of AWS and some of the benefits we see from the AWS, maybe it'll be useful to just understand what's in the future roadmap to further strengthen the agentic AI solutions or solutions supported by AWS.

Divakaran Ullampuzha Mana

Now, I'm really excited about this because we are at an inflection point, right? The market is projected to reach around $52 billion by 2030, and the Gartner predicts that by 2028, at least 15% of the work decisions will be made autonomously by agentic AI. So, our vision, AWS's vision and philosophy centers around three strategic pillars.

The first one is differentiated innovation, right? We are not just providing the infrastructure. As I said, we build co-developed advanced agentic solutions.

And all of what we build, a vast majority of them are built working backwards from customer problems or customer use cases that they are trying to solve, which means we are collecting all that information and building it and hence has a much more likelihood of solving most of the problems that a customer would face as they design and build agentic AI solutions. Second is our next generation data foundations. As we had discussed earlier, agents are only as good as the data they can access.

And our next generation Amazon SageMaker with Unified Studio, open lakehouse architecture combined with Amazon S3 vectors, reduce the storage cost by a large amount while maintaining subsequent performance. And this creates a great foundation for truly intelligent agents as well.

Third is our advanced infrastructure optimization. Our custom AWS design and developed GPU chips, Trainium and Inferentia delivers a much better price to performance with auto scaling capabilities, which means again you can start very small, grow the usage of GPUs as your workload requires and scale down as well if it is not needed. So, this makes agentic AI economically viable for organizations of all sizes.

Our philosophy is very simple. AWS wants to be the best place to build and deploy the world's most useful AI agents. And we are balancing rapid innovation with security, reliability and operational excellence that enterprises demand.

Gautam Singh

Well, that's fantastic.

And I can't wait to see what happens as time unfolds and AWS introduces all these new propositions.

Divakaran Ullampuzha Mana

Gautam Singh, you are at the forefront of agentic AI implementation across industries. Can you tell us what the future roadmap looks like for agentic AI services delivered by your business?

Gautam Singh

Sure, Diva. Thanks for asking that question because I am equally very excited when I look at the roadmap ahead of us.

For one simple reason, we as WNS, the wider WNS, and I'm WNS Analytics within WNS, but the wider WNS is a BPM domain focused, a domain depth company. And for the last 20 years, we've been managing business processes and transforming them as WNS. And now if I look at, as I mentioned right up front, the power of agentic AI, all of those processes and more can be reimagined.

So, the workload and the opportunity for us as WNS Analytics is just very exciting. But if I just double click on that, you know, where all is it going to be really game changing? So the simple answer to that is workflows or processes that are highly human dependent today, contact centers, back office processing, where lots of human interaction or human work, human tasks need to be done to look at contracts, read documents, to mine documents, take data from one document and another, put it into a new place, etc. Those processes, in particular, the more complex end of those, because the non-complex ones have already been automated.

The more complex end of those workflows is where the real opportunities are. And these include, I mean, just as a few examples, clinical summarization for our pharma and healthcare clients. There's a lot of time, effort, and human cost involved in managing those processes.

And those can be reimagined, where the human plays a different, perhaps more value adding role than with a lot of the work, the underlying routine work being done by agentic AI preparation. There are contact centers, as I already mentioned. There are also, frankly, I would say, business processes, which can be more effective, because now you can incorporate unstructured data into the consideration of the workflows.

And those include, for example, investment research. Where do we invest money? Look at those processes. We can now incorporate unstructured data.

To give you an example, if you're an equity analyst looking at retail stocks, up to now, you were looking at financial savings, historical data on how a retailer is performing. But actually, with the agentic AI solutions, you can feed in satellite imagery on how full the car park retailers are, which will give you a predictive upfront view of what their future sales are going to be, as opposed to a historic view of their financial savings. Now, if that all gets incorporated into an equity analyst's evaluation of a particular stock, that's different.

That's game changing. And by the way, it no longer needs to be done by humans. Agentic AI solutions can take on a bulk of that role.

So, these are just some examples. But fundamentally, I'm excited, not because we can tweak, we can re-engineer, but because we can reimagine a lot of the work that we are doing for our clients already today as developers. I think this has been a very interesting conversation, guys.

If there are any closing comments from your side, if you want to summarize from your side, let's do that. And then I can perhaps close this as we are coming up to our roughly 45 minutes.

Raffaelle Breaks

Yeah, I think the summary here is to your point around reimagining processes and platforms like SKENSE really present an opportunity for organizations on their AI transformation journey to rethink their data and really unlock value from that data that's currently being maintained in many different places, many different formats, and aiding them in really that transformation so that when they start to implant more agentic AI into their organization, they know that data debt is not going to be a problem.

So, just very excited about the opportunity, very excited to keep developing SKENSE with our AWS partners and bring these solutions to market with you.

Divakaran Ullampuzha Mana

Yeah, I am as excited as well. And as you reimagine your processes, as you develop SKENSE and other platforms using agentic AI, infused agentic AI, AWS is providing you everything that you need, whether it's starting from the infrastructure all the way up to the ready-made agentic AI applications that you can use to build something that is scalable, secure, and trustable, and can meet all of your customer requirements as well.

And we are happy to partner with WNS and the SKENSE team as SKENSE evolves into a fully agentic AI system.

Gautam Singh

Thanks, Diva. Thanks, Raff.

I think I'll just summarize by saying we, and I hope our listeners find this very insightful and equally exciting for themselves. But I would summarize this as saying, we're just at the start. Already, there is so much change happening through agentic AI.

But if I look at the investments we are making in SKENSE and other platforms, the investments we are making are relooking existing business processes to see how they can be reimagined. And we look at what our partners, in particular with AWS, that we have a strong partnership, what you have in plan for the future. What we've got today is just a start.

I think when we have this conversation, perhaps again in a year's time, there will be significant progress being made. And it'll be even more exciting then to think about what the future looks like. But for me, the future is very exciting, very fast-changing, and actually going to be very beneficial for everybody, both the users of these agentic AI solutions and those who are building them.

Thank you very much.

Raffaelle Breaks

Thank you so much.

Divakaran Ullampuzha Mana

Thank you all.