Hello, everybody, and welcome to CNF Talks, the series in which I get to speak to a forthcoming speaker at one of CNF's upcoming conferences. Today, we're looking at the Agentic AI for Leaders Summit, which is being held on the 17th of September in central London.

And I'm delighted to have with me Niresh Rajah, who's the Chief Data and AI Officer. He's Chair of UK AI for Financial Services, a Fellow at Imperial College, and an Associate Fellow at Oxford University. So extremely well qualified to answer our questions, and Niresh is chairing the event itself.

Hi, Maurice, good to see you. 

Very nice to have you with us. 

There's been a lot of media discussion about agentic AI, and the sense that one gets is that many companies have been experimenting. But how do we move from experimentation to value? 

I mean, that's a great question, and one I advise a lot of boards, and one I'm asked quite frequently in terms of proof of concept into value and creating return investment. I think there are a number of factors that companies need to look at.

The first is selecting the right use case. And that's important in terms of what is the impact on the customer? Are they looking to launch new products, new services? And how is AI going to be augmenting the workflow to be able to provide better services to clients or actually becoming even more operationally efficient? But picking that right, the correct use case is, I think, number one. The second is really looking at desirability and feasibility.

And what I mean by that is, is the use case truly? Can we do some mapping to understand what is the return on investment from this use case? And how do we measure value? So that's the desirability part. And then the second part, then, is feasibility. Can we do it? And you might be aware that 70% of AI projects do not create a return on investment in terms of some of the analysis.

And 30% of AI projects are canned after proof of concept. So the second part is really important in terms of what I'm articulating is check the feasibility. And then the feasibility really comes to, can we do this? And how are we going to implement this? And do we have the ability in terms of both the people, but also the AI, and importantly, the data that fundamentally provides the key component? Can we make that work? So those are some key elements around moving from proof of concepts into scale.

Very interesting. Turning now to the governance of autonomous systems and the responsibilities of boards. At our recent flagship event, City Week, we had the head of the CFTC come to speak.

He's talking about facing the question, really, from a regulatory point of view. How do you regulate an autonomous AI agent working on chain using a digital wallet to make money, learning as it goes, and changing strategy? And I suppose taking that and thinking through from any sort of corpus point of view, how do boards and executive teams govern autonomous systems responsibly? 

Indeed, look, I've been a material risk taker in financial services as a chief data officer. And it is the first part, really, I think everybody now understands is the accountability sits with the executive, or the senior leader in place.

So we can't absorb that responsibility or accountability just from having agentic AI automated systems. I think there are a number of things, a number of ways that executives and boards can really make sure that there is some assurance and oversight. The first is actually picking the right, select the right parts of the automation, or right parts of the workflow, and making sure that you're picking that very sensibly.

So not everything can have agentic AI workflows implemented. So are you picking that with real clarity? And are you weighing up the opportunity versus risk? And we haven't yet talked about responsible AI and AI governance. But I think those are really important facets of are you really clear why you're selecting this process to be identified? Second, I think the human in the loop is sometimes a term that is overused.

But fundamentally, it's how much human intervention are you placing into that workflow? And within that workflow, are you able to monitor when things are within guardrail tolerances, and when they fall outside of your guardrail tolerance. So the level of monitoring and oversight has to have intervention, but also needs to have MI that flows to really indicate when things are not working as they should do. And I think that is important to design systems in a way that happens.

And the third part is, I think, really, you need, as I see firms that are successful in creating agentic AI, and I'm advising some of these is take a step by step approach. So we talked previously about scale. And I think the scale part around specifically when you're creating agentic AI systems, is actually increasing volume and increasing bandwidth around agentic AI in a more methodical manner.

So it is not a big bang approach. And hence, you're able to see some of the outcomes that are being created in a step by step. And it is not everything, you know, you're not putting all of your organisation at risk through agentic AI.

I think those are three practical ways I would recommend firms actually approach agentic AI systems. 

And I suppose sort of slightly flipping the question and looking at it from the customer point of view as opposed to responsibility and boss, how do we build trust and compliance and accountability, I suppose, into agentic workplace? Because I suppose there is a danger that if there are some very unsatisfactory outcomes, potentially, that that might undermine consumers, for instance, consumers trust in responding to agentic AI. So how do you provide them with that kind of assurance that you are compliant?

And Maurice here, I'll probably think about customers, but also employees, I think they're a flip of a coin in terms of thinking through the trust angle.

Some of the boards and senior teams I'm advising, I think I really talk about creating a revised AI strategy. Now, whether that's part of a business strategy or not, I think it doesn't really matter. But that AI strategy needs to be very clear in terms of AI ethics, data ethics, responsible AI and AI governance, you know, some of the things I've already referenced.

So we're really clear in terms of what we will use AI for and what we will not use AI for. I think that's a really important part. Secondly, we will really clear in terms of actually the data set in terms of going back to data protection, data privacy, how we will use customers and clients data in terms of the AI workforce or AI mechanisms and what we wouldn't use our customers data for confidential data specifically.

So I think we need to be really clear on those on those elements. Then the second part then here is when we say we're not going to do anything, then there is actually having a very clear risk and governance component around how are we actually going to match what we say we're not going to do. And so all AI use cases, I chair some AI steering committees in some organizations, I think that is important in terms of actually keeping on top of the value obviously that AI is going to create.

But secondly, making sure from a risk appetite standpoint and risk thresholds, you're challenging those use cases so that you don't deviate from the things that you said from an AI ethics and responsible AI that you won't do. I think that builds a lot of trust. And the third part is education and literacy.

And that literacy extends both to customers to let them know how to use AI and where AI is beneficial and how we're actually keeping in line with what we said we would do. That AI education also extends to employees so that they really understand in terms of how AI is being used. And it's not just about replacing people in terms of what that looks like.

I think those are three pragmatic things organizations can do to build trust but also sustain trust.

Absolutely. Final question because I think we're pressed for time now.

How do businesses turn agentic AI into a sustainable source of competitive advantage? 

I think we're still, I think all organizations are still figuring that out. So I think we're very early in that process. But in my view and some of the thinking I'm helping boards do in this space is really think about the competitive market space, the changes in the business model.

So one of the things both in regulated sectors and non-regulated sectors is AI now poses a threat in terms of the business model. And that happens in a number of ways. How are boards and executive teams challenging themselves in terms of the changes to their business models and the competitor marketplace in terms of that? And then obviously the regulator space when you're in regulated sector.

So that type of analysis is really important. And then I talk quite often about a model in terms of, I call it a 721 model in terms of actually thinking through what type of AI use cases are for the here and now in the next 12 months in terms of efficiency and productivity. The second part in terms of 20% of your portfolio is really thinking through new AI augmented products in terms of whatever components.

And the last 10% is actually disruption. How are you going to be disrupting the market with AI augmented products? And what is the firm of the future in terms of whatever industry you are in actually look like in three to five years with AI? And I think that type of analysis has to be done, especially around thinking at board level. Very interesting.

Well, Niresh, sadly, we have run out of time at this point, but thank you so much for joining us today. And for our viewers, do have a look at our website, www.cityandfinancial.com for further information about this event, the topics we're covering, the speakers that we have, and obviously to register. We'd love to see you at the event.

And Niresh, we'll see you there on the 17th of September at the summit in London. Thank you so much.

Thank you, Maurice. I'm looking forward to chairing the event.