EXCLUSIVE: “All the Gen” – Jonathan Hall, NatWest in ‘The Fintech Magazine’

Thanks to ChatGPT and Bard, consumers are more familiar with – and perhaps more concerned about – this generation of AI than any previous iteration. That means banks have to tread a very careful path, says NatWest’s Jonathan Hall

The institutions we trust with our finances had their technology toolboxes filled long ago with AI and its subsets, including machine learning (ML) and deep learning (DL). And, on the whole, there has been little consumer interest in how those technologies were applied in banking.

We’ve simply been happy to benefit from the results, with faster credit decisions, helpful budgeting apps and personalised offers and pricing. Generative AI is a little different. The public has come to know a lot about that in the past 18 months – and not all of it good. How therefore should banks approach it?

Take UK banking giant NatWest.

It recently announced a collaboration with global technology developer IBM using generative AI to develop Cora+, a substantially upgraded version of the virtual assistant many of its customers are already familiar with. With a GenAI reboot in the background, Cora is now designed to respond more accurately to customer inquiries. NatWest is also expanding its partnership with Amazon Web Services (AWS) to accelerate the use of AI – including AWS’s generative AI solutions. – to drive towards its goal of helping 10 million people manage their financial wellbeing by the end of 2027.

“We already deploy AI right across front, middle, and back office, in a whole range of use cases,” says Jonathan Hall, NatWest’s head of digital for commercial and institutional business. But it will act with caution as this new, improved version with its apparently limitless applications fast develops, he adds.As with any AI, Hall says, the use case is not the constraining factor.

“It’s about an organisation’s aptitude for the technology, it’s the capacity to then deploy it, and its risk appetite to do so.”

The areas where artificial intelligence in all its forms can most help the bank and its customers remain the same.

“One is personalisation, and how we can deploy for customers, to really give that relevant, hyper-personalised experience,” says Hall. “The second is around protection. That largely is about how can we improve our risk management capabilities: we can get more predictive, in terms of some of our forward-looking risk assumptions.

“The third area is around ops, just increasing optimisation and digitisation, efficiencies, reducing cost.

“Then the final one is around workforce augmentation and knowledge; how we, internally, can deploy AI. Because we’ve so much data, spread across so many different franchises, pooling some of that, and making it really accessible for some of our teams, is also just a huge, huge opportunity for us.

“So, AI and GenAI is exciting, but we’re taking an approach that I’d describe as mindful innovation with safe exploration. We are going to be very diligent and choosy about where we play in this space.”The bank certainly isn’t ready to let GenAI operate with any degree of autonomy.

“The majority of the things that we’re looking at with the new technology we term ‘co-pilot’,” says Hall. “That’s a less risky way for us to deploy this. That means it’s not direct to customers. We will deploy it side by side with a relationship manager, or a web chat agent, or a telephony handler, so it will actually draw information forward as it analyses a web chat in real-time, for example. That will help speed up interactions with customers, and then we can test the outcomes with our specialists.”

So where does Hall see AI making a difference in core banking tasks like, say, ensuring the accuracy of collateral valuations or in risk management?

“Traditionally, what you’ll find in banks is that the credit processes don’t talk to each other,” says Hall. “We might do some scoring up front, but then we might hold the security for customers on a different platform. And while we might have some very good descriptors about what security we hold against those loans, using AI, we can pull in different data sources, use the unstructured to structured type principles that this technology provides, to give us much better real-time valuations and assessments. That means a customer may be able to draw down further against that security. It might also have an impact in the back office on our risk-weighted assets – how, for instance, commercial real estate valuations are changing over time.”

“We’re taking an approach which I’d describe as mindful innovation with safe exploration. We are going to be very diligent and choosy about where we play in this space”

When it comes to risk management, ‘the move from static to real-time early warning indicator systems is really what the opportunity is‘ continues Hall. “As these systems can pull in lots of different data, you can imagine dashboards that could change, red, amber, green, on certain parameters. We have some of that stuff deployed in certain areas. Is it pervasive across the whole risk management framework? Not yet. But I can absolutely see an environment in which we end up in that position.”Ensuring the integrity of data that is being compiled from a vast array of structured and unstructured sources is,

of course, paramount for any business. And it’s an area that Hall says has been revolutionised since the early days of databases when the industry adage was ‘garbage in, garbage out’.

“There are programmes now which can clean up data for you, as it goes through. So, for example, it can identify where there are duplications, it can identify trends, and where there are particular datapoints that sit outside of those trends,” he says. “So, actually, the coin has completely been flipped, in that respect.

“However, there’s a kind of further consideration, which is, as the technology is increasing in complexity, and is advancing, then, as a bank, we need to ensure that the models are transparent.”

Another key use area for AI in financial institutions is in the constant battle to combat fraud. AI can help identify fraud via transactions and/or movements in a vast ocean of data by detecting and flagging unusual patterns. Machine learning can therefore train AIs to become the bloodhounds of the global financial streams. Hall acknowledges that NatWest is keenly aware of the issue, for which it calls on third-party specialist help.

Continuing, he says: “We have got to continue to upgrade all of our systems, and help protect our customers, so we’re constantly scanning the market to see what the best-in-class opportunities are. We absolutely need third parties to help us solve some of these problems.”

A further area getting close attention from all who use AI in customer-facing industries is how to comply with the UK’s recently introduced Consumer Duty regulation, which stipulates that firms must design products and services that aim to secure good consumer outcomes. Interestingly, Hall makes the case that AI can be used to ‘self-police’ products to ensure they comply with Consumer Duty.

Explaining the process in detail, he says: “We can start deploying this technology to look at far bigger datasets than humans would be able to get through. But we can also bring in customer feedback. We can take transcripts of calls, etc, and then analyse those to get a real sentiment, to make sure, number one, that the customer has been able to make an informed choice, that our due process and frameworks have been followed, and we can then combine it with the data from the output to make sure the customer has gone on and used that product as it’s been designed.

“There’s a whole range of examples and use cases where we can start deploying some of these technologies to evidence that we are being compliant.”

For now, though, the closest NatWest customers will come to experiencing generative AI is with the new Cora.

“We’ve created a closed large language model, based on our website, and some other information,” explains Hall. “So, instead of having to search through pages, a customer can, in a very conversational way and with a prompt – which we’re probably all used to now, by working with ChatGPT or Bard – have Cora pull it forward for them in a much more efficient way. That’s very exciting.”

This article was published in The Fintech Magazine Issue 31, Page 66-67