Paul Morrison

Hello and thanks for joining us on Retail and Consumer Pulse brought to you by WNS. In this podcast, we explore the world of retail and consumer goods, looking at the latest and most innovative ideas from industry experts and leaders. My name is Paul Morrison. I lead the WNS retail and consumer practice in Europe, and for today's session, I am delighted to be joined by Leigh Varnham, Chief Growth Officer at WNS-Vuram.

Leigh, tell us about your role at Vuram.

Leigh Varnham

I head up the commercial function at WNS-Vuram for Europe.

And essentially, what we do as a company is identify hyperautomation solutions for some of our existing customers and some of our new customers as well. So, how do we find and implement solutions for customer problems? I spent the last 15 years in the automation arena both with hyperautomation software companies like SS&C Blue Prism and Pega and organizations like WNS who take these products and implement them.

So, what is Vuram? Vuram was acquired by WNS a few years ago. We are a pure play hyperautomation organization. We focus on a range of different technologies, including IDP (Intelligent Document Processing), RPA (Robotic Process Automation), orchestration, process mining, process intelligence, and of course, what we're here to speak today - about agentic AI.

Paul Morrison

Absolutely! So, a great guide through this topic we've got today - agentic AI indeed.

Let me briefly frame what we're seeing with this amazing trend in the market and some of the commentary. For example, McKinsey said 60% of companies are exploring AI agents in their customer engagement. BCG said 67% of companies are considering AI agents in their transformation journeys.

And looking at Google Trends, there's been a hundredfold increase in the search term ‘agentic AI’ in the last year. So, there's a massive tidal wave of interest and excitement about agentic AI. And in many ways, I guess from my perspective, this is the follow-on from that wave of enthusiasm two years ago around generative AI.

That leads to our first question for you today, Leigh. What is agentic AI and how does it differ from other types of AI?

Leigh Varnham

The primary difference between agentic AI and AI lies in a few different areas. One is autonomy. Then there is decision-making and behavior. So, let me give examples of what it isn't.

If we look at some of the examples of AI that I think a lot of people can relate to - virtual systems like Siri or Alexa are what we would call AI, or recommendation systems on streaming platforms like Netflix and Prime. A chatbox is another good example of AI, and they normally, not always, follow a predefined set of instructions or algorithms without true autonomy.

You know, things are changing even as we speak. You probably saw in the news that Alexa or Amazon is going to be bringing out agentic AI on all their Echo devices, starting at $19.99 a month and only available in the US at the moment. So, things are changing, but if we look at agentic AI, it's an advanced subset of what we would call AI.

And going back to my earlier point, it's capable of making decisions on its own. So, it doesn't need direct intervention or constant human input in order for it to function. That independence allows it to take actions and solve problems without that kind of external control. And it's very much specific for goals. Agentic AI systems are purpose-driven; they understand objectives, and they understand how to strategize them. It might be anything from optimizing a business process to performing complex tasks like navigation in your Tesla or an EV car.

They can also learn. You're not constantly changing the logic or changing the way that they're learning. So, they've learned by themselves, and they're able to make quite complex decisions. They go beyond reactive tasks that you would normally expect them to do. And the last one, which is probably the most interesting, is the way they can interact with the environment. If you think of the rise of IoT devices being able to take all this information both from a static point of view, business rules or ethics, but also take care of things that are constantly changing, whether it's temperature or humidity or the number of revolutions, and then make those decisions very quickly based on all that data.

Paul Morrison

That's great! There's a load in there to unpack, and I guess one of the key themes you've touched on is the fact that there's interaction. It's building on the interactive capabilities of large language models or generative AI – that autonomous dimension around taking decisions, acting on them and following them through. And I think, following on from the fact that means agentic systems can target more complex tasks, perhaps, more than previous waves of automation have been targeting. So, I think you know all of those things taken together, that's a very broad canvas.

Just to add some color to it, I'm thinking through other different types of agentic AI that we should be thinking about, from your perspective, different ways of thinking about the range of agentic AI providers out there.

Leigh Varnham

I think, there are pure play agentic AI companies that are looking at this from a particular line of business, vertical and problem area. And then, those organizations probably, mainly SaaS companies, that are having to kind of bolt on retrospectively, apply agentic AI to their product. And I think, to a large degree, if you look at some of the providers out there, and I'm not going to name any names today. But the kind of drive is, first of all, how do we make our product better for our customers using agentic AI and those features and functionality that we just talked about, but then also, can I drive the flexibility, usability, time to market of developing some of those solutions?

So, what we're saying is,, it’s not just the rise of organizations or software companies applying agentic AI to the way their users, organizations and third parties build applications, but also how they interact with end-users. So, two broad markets – those that are very specific and those that are now, I guess, running to make sure that they apply agentic AI into their products.

You would have seen some of these adverts. There's a very famous advert with Idris Elba, who's a spokesperson of one such organization. Microsoft is growing and getting into this market very quickly. Some people are actually talking about the death of the GUI, right? So, instead of the kind of legacy, where humans used to write the code, type code, you now have, or you have the ability to use, agentic AI to create applications. So, whether we need a GUI is the question.

Paul Morrison

Yeah, that's a really good one. I think, what you've said is the most interesting or important question about the development of agentic AI is that trade-off or that spectrum ranging from the pure play to existing core applications. It is striking when you look at any leading software provider at the moment. Many of them have already implemented or started to share their agentic functionality. And at the same time, there are many digital AI native players, like Open AI, for example, with an operator providing a sort of the standalone . At the heart of it is this question: what is the work to be done and thinking about the work of an employee or digital employee?

It's all about the workflow. What is the sequence of tasks to be done and is that best delivered by a specialist platform like SAP in finance or WNS TRAC One-F in finance, or is that going to be augmented by additional agentic solutions as well? It's going to be very important and interesting for organizations to work out how that plays out in the future. There are loads we can unpack, but this takes us on to a question about use cases for retail, consumer goods and manufacturing organizations? Where are some of the early use cases, and where do we think this is heading? Do you have a perspective on that, Leigh?

Leigh Varnham

So I think, there’s the opportunity. We can start naming some of the names that are using agentic AI or AI with inside their own industry for certain things. So, I'm sure that everybody has seen. I've actually been into one of these Amazon Go stores in near Wembley. That was my first encounter. And anybody that doesn't know, but you go into the shop, and you don't have any cashiers.

The one I went to actually had a security guard, but apart from that, there were no cashiers. The sensors and cameras track the products, what I'm putting into my basket or what I'm putting into my pocket. They can also see when I put it back, and I just get charged for what I pick up and what I don't put back. So, that whole concept, the horrible concept of self-checkouts or scanners or having to interact with people and checkouts, that doesn't apply anymore.

So that's one example. I guess the other one, with Amazon again, is if you go into their website, of course, there are full of personalized product recommendations. It's always interesting from my perspective to see what my other half's been looking at, and what Amazon is now predicting that I would like to buy. Sometimes, I do feel that she does it just before Christmas and birthdays, so that's another use.

If you start looking at manufacturing, you got Tesla, who actually have robots in their factories. It's been going on for some years with different providers and different car manufacturers as well, obviously speeding up production and reducing human error. You've got BMW. They're using it for quality control without having to get people there looking at all the parts and all the panels, that are 1coming down line automatically that AI can actually define and understand whether that is fit for purpose or not.

We've got General Electric. They're using it in manufacturing facilities to optimize supply chain. So, if you remember the good old days of just in time when organizations were always, you know, they did have a whole load of stock. They wanted to order just in time, and then you obviously have things like the Suez Canal problem. And you then start using AI to actually predict what might happen both from a macro and micro standpoint, but also to predict when you're going to need certain raw materials. And so, you can have that kind of constant production without any manual intervention, but without having huge amounts of stock as well.

Paul Morrison

Some really good use cases there, and I think, the thing that's striking is they are scattered across lots of different parts of the value chain. So, we've got customer experience, we've got supply chain. We've got another sort of great example in around HR. I think, a number of really strong use cases there. A few of the ones that jumped out to me, on my radar is Walmart using agentic around predicting demand and stock replenishment processes, P&G around analyzing customer data, website visits and purchase history that feeds in down the workflow to tailoring marketing messages.

Paul Morrison

We have P&G using agentic to analyze customer data in terms of website visits and purchase history and feeding that down the workflow into how marketing messages are tailored for potential future business. Then we have Sephora using agentic around personalizing beauty advice and customer queries through virtual agents, following workflows and interactions from initial queries through to sale and an after-sale. So, some very interesting use cases out there.

Let's turn to some of the challenges we've heard about what agentic is and can be and where it can be applied? What are some of the issues that you see at the moment, and maybe, looking into the future?

Leigh Varnham

While agentic AI offers lots of benefit, I think, there's still some questions. There always is with this kind of new technology, and some of these, and probably most of these and anything new in my view.

So, I think, one of them is certainly about ethical concerns and accountability. If it makes a decision, and that decision is wrong for whatever reason, whether it's in healthcare, whether it's driving a car, who is actually at fault? Is it developer? Is it AI? Can't really blame agentic AI. Is it the organization? And we've seen this quite a lot. If you remember, when Tesla originally kind of started talking about the self-driving car, then the question there is for car insurance - if that car is involved in the crash, and it's self-driving, is it Tesla or is it the person behind the wheel? Is it something else?

And if you look at some European countries over here, we insure the person. Sorry, we insure the car, not the person, whereas over in a few European countries actually ensure the person. So, you know, is Tesla now going to have to have in depth insurance, for example, for these kinds of, and you know, we've seen a couple of those accidents happen already, but not in this kind of self-drive manner.

So, when cars do become autonomous, who is actually ultimately responsible? Then the second one is bias. And we've seen it with AI in the past. At what point? We've seen it as well. I think, if anybody's been kind of reading, look at the news recently with the DeepSeek agentic AI that came out. That substantially cost less than ChatGPT. If you start asking it anything to do with Tiananmen Square or anything like that, it didn't give a response and refused to give a response. So, you can see where some of this kind of bias starts to come in.

Security, privacy risk needs a lot of data in order to operate. And can you somehow start to challenge and manipulate it? I've already read about organizations that are manipulating certain AI models or LLMs to promote their product or service. So, I go to ChatGPT now or another AI model and go - Who's the best producer of televisions? Who's the best producer of cars? And there are organizations out there that are basically feeding these models with the right answer, which is their organization or the organization they're working with.

Obviously, the big one is job displacement and economic impact. You know what's going to happen. We've seen it before, during lots of different revolutions right throughout the years. We heard about RPA and automation. Has it come to fruition? No, not really. But there is that fear around. Can this actually lead to reduction in jobs? Do we and can we retrain some of those people? You then get the kind of view, well, if some people can use AI and some people can't use AI because some are older, some are less literate in terms of understanding technology, is that then creating a disbalance or inequality between the two?

Paul Morrison

Yeah.

Leigh Varnham

And then the big one that everybody is talking about, which is regulatory and legal, and lots of governments all met in Paris very recently to discuss this. I think it's going to be really hard for, as we’ve seen in Paris actually, for those organizations and countries to come to an agreement on this. So, I think regulatory and legal issues is going to be a big one going forward.

Paul Morrison

Apart from that, it's all very clear.

Paul Morrison

It's some really big topics there to chew on, and we can only scratch the surface today. I think on the first one around the responsibility, liability question, that is a big one or a big fear. When we're talking driving cars and operating heavy machinery, the risk is clear and present. But even if we're thinking about software-focused agentic AI and automation, then there is a challenge, particularly, as we see the defining feature of agentic AI is this ability to work autonomously, that is, for it to observe, to assess, to make a plan based on an objective, and then to enact and to deliver that.

So, the risks and the impacts of that will vary depending on the activity in question, whether that's onboarding a new staff member or dealing with an irate customer. The impacts of that independence will be different. But I think, that will obviously need to be worked through in great detail.

The Employment and Society one is a big one as well. I think just to flip that on its head, there's a question about agentic - if one end of the spectrum is these autonomous decision-making AI expand and deliver huge swathes of work is one end, the other end is more grey. And if we look at today, whilst we have set out, lots of use cases and lots of early adopters where there is success. We are still right at the beginning of the adoption curve for this technology, and there have been technologies in the past, where the growth hasn't always met the hope or expectation. And so, I'm wondering, is it different this time with agentic AI that they're the power of large language models? The ability for that to drive adoption of autonomous AI across workflows, is that a no brainer? What do you see some adoption challenges as well, Leigh?

Leigh Varnham

I certainly see adoption challenges. I mean this is crystal ball time, right? And there's lots of permutations. And in reality, lots of people have tried to predict the future in the past and very rarely do they manage to get it to 100%.

It is a fundamental change. Organizations, software companies, the market, the analysts have been talking about AI for decades, and having the power in the hands of normal people. So, people can actually use this is what I think is a game changer.

So, it's going to be interesting in terms of what happens. I think, one of the things a lot of Marvel fanatics will know is this quote, but it wasn't actually firstly uttered by them or the Marvel Universe. It was actually Voltaire - With great power, comes great responsibility! And ultimately, there is a great amount of power that you could wield with agentic AI.

One of my former bosses actually talks about this idea of Sentinels or guard rails. So, how do you put some parameters around everybody being able to use this life-changing technology? Things are going to get quicker. They're going to get faster. They're going to get easier to do. And that means certain things that organizations don't want you to do, you will be able to do.

If you start thinking about the past, history where people have created a swarm effect, whether it's to change stock pricing or stock prices or do other things, it just requires something like that and using agentic AI can really help with that. So, it's going to be an interesting ride for sure. I think, it is substantially different and very materially different from AI.

Paul Morrison

It's very interesting, and we've never had anyone quoting Voltaire on this podcast before, so that's the first. I think that's really interesting. And I subscribe to that view - that in the balance of probabilities, the impact will be broader and accelerating. Just having voiced that question about adoption, I think that’s one of the things that's interesting about the technology and what it promises. Because of the use of large language models to enable interaction, that can be interaction between an agent, a human, an employee or a customer. It can be between the agent and a core system or a website, and it can also be agent to agent as well.

So, there's this whole new space of agentic AI interactions that becomes possible very quickly in a way that wasn't before and all previous phases of automation. If we think back only a decade or so to the growth stages of RPA, a lot of the ideas around RPA or a robot for every person, around the transferability of robots to different parts of a business and the infinite scalability, a lot of the ideas around that actually resonate quite strongly with what we're seeing in agentic AI. But with this new technology layer, it seems some of the implementation scaling problems of RPA, for example, is one of many automation technologies that is sidestepped.

So, if that is the case then the scaling of this becomes faster, and it comes back to the individual, in our case, retailers and consumer goods companies in the next three to twelve months thinking through how this fits in their technology, roadmaps and their digitization. What your thoughts are in terms of advice to organizations that are thinking through, is this relevant for now?

And I think as a sort of bare minimum. I think it's important for retail and consumer goods listeners to think about their current technology footprint in an area like finance or HR, procurement, supply chain, understand their core applications at the moment where some of the use cases are in that, and understand what your current technology providers have mapped out, and how are they thinking about the potential here. I think some of this will build on previous discussions and ideas from generative AI to which it is connected.

I think, it will be continuation of that, but if we're thinking about AI that is more autonomous and can work across a workflow, you know what might that mean for how we deliver HR or customer contact? So, think through how your existing partners and technology providers are talking about and thinking about the future. I think as a first step, do you have any, before we close, any sort of key actions for the next few weeks and months?

Leigh Varnham

Yes, it's a bit like with RPA. It's really easy to look at what you're doing now and then transpose that and convert that into how you would use agentic AI. The really interesting stuff and the really hard stuff to do is then taking something you're not doing right now, that kind of blue-sky thinking, and then thinking how you can apply to it. That is really difficult because you need technically understand what agentic AI can and cannot do and whether it's possible, whether you have the data, and also understand the process. That's really hard for any organization to do, and that's where you need experts, who actually understand what the capabilities are, what the limitations are, and what the strengths are . So, I think that certainly is key.

The other thing as well is RPA was very good at, I guess tactical, very task-ßspecific automations. And agentic AI, to a large degree, has kind of taken that over. But the great thing about agentic AI is that you can very quickly and easily create these independent, autonomous ways of working, and then you can connect them very easily as well. So, you can have a whole load of, let's call them agents, AI agents sitting in HR, a whole load of AI agents in finance, a whole load of AI agents in sales, a whole load of AI agents in customer service and they can all talk to one another. That is the great thing because you're not just building one thing and then you stop. It's about then building others and then connecting these together. So, you really start to maximize whatever you're trying to achieve, whether it's cost saving, whether it's customer experience, whether it's speed. It can be a combination of all those different things, but essentially, you are building for the future, you're not just building for now.

Paul Morrison

Very nice sound bite on which to end there. So, building for the future! Leigh, really appreciate your time and expertise. I think we'll need to stay close and check back in before long because this is a super-fast moving area. But thanks very much for your time today, Leigh.

And thanks to our listeners for joining Retail and Consumer Pulse today. If you've enjoyed the show, please do like and follow us and stay tuned for our next episode. So, thank you and goodbye.