AI and Banking Intelligence

How’s your behavioural intelligence unit? While AI agents are grabbing banking headlines, there are interesting developments happening down in the more prosaic areas of transaction data and fraud detection. Is it notable that Nvidia is helping the UK’s first digital banks to develop new models of banking intelligence? The chipmaker is helping banks to build “intelligence layers” that promise to replace multiple siloed systems in banks and turn them into single pre-trained transaction models.

Starling is among the banks working to build internal AI-driven financial crime investigation units. Starling says its new model will be trained on proprietary data from over 15 billion anonymised transactions to give it deeper insight into the needs and behaviours of its customers. “The model is set to go into production in early 2027, initially focusing on services around fraud detection and then to augment the AI Starling Assistant,” writes Finextra. “Starling developed the model with Nvidia's financial services team and is working with UK cloud provider Civo to ensure that that the model will be trained and hosted within a sovereign environment and that all data is within UK jurisdiction only.” The UK government is among governments wary of reliance on a small number of cloud computing providers such as Google and Microsoft. An outage at Google last week affected a number of UK banks including Revolut, Lloyds and HSBC.

Build Your Own Transaction Foundation Model Blueprint by NVIDIA | NVIDIA NIM

Nvidia has also been working with Revolut to build Revolut’s own Transaction Foundation Model using its history of customer transactions, called PRAGMA, or Pre-Trained . (Indeed, Nvidia made a $196 million investment into the digital bank. If you want to read about the Transaction Foundation Models in detail, Capitec CIO Andrew Baker has written about it: “Somewhere inside your bank, in the time it takes to read this sentence, a few thousand payments have arrived that nobody will ever look at. Each one carries an amount, a merchant, a timestamp, a channel and a counterparty, and each one, considered on its own terms, is entirely unremarkable. The information that would tell you whether to worry about any of them is not in the payment at all. It lives in the relationship between that payment and the several thousand that came before it from the same person, in the rhythm those payments have settled into over four years, in the small drift of the last six weeks that nobody has noticed.”

To train this model, Revolut generated 24 billion transaction events from anonymised and tokenised records from its own customer base to build a model of customer behaviour that can then be used to serve several “downstream models”. Let’s return to the financial crime and financial detective analogy. As Baker notes, a retail bank typically has six teams separate models, one model doing card fraud, another doing credit scoring, and so on. Imagine each team, says Baker, as a detective. “Each detective develops a private filing system and a private set of instincts, none of them benefits from what the others have learned, and when a new category of crime appears the only available move is to hire another detective who starts from nothing. PRAGMA is an attempt at the alternative, which is to train one detective who has read every case file in the city and then give that detective a short briefing for each crime type as it comes up.”

Now equipped with a UK banking licence, this week Revolut announced the launch of its consumer lending offering in the form of unsecured personal loans in Britain, its UK chief executive Francesca Carlesi told Reuters, taking Europe's most valuable startup into a lucrative consumer lending market dominated by the country's biggest banks. “The move pushes Revolut beyond its core payments and subscription products and into a riskier business as it seeks to compete more directly with lenders including NatWest, Lloyds, Barclays, Santander's UK business and HSBC. However, Revolut’s chief executive Nikolay Storonsky has said that he wants to build a no-risk bank, pledging to keep lending to around 15 per cent of the bank’s income. Simultaneoulsy, Storonsky signalled that the bank has ambitions to grow into a major technology company that can compete with the biggest tech players in the US.

HSBC shares dropped as the bank narrowly missed targets, reports Reuters. “Analysts noted that the bank's net interest income slightly missed expectations, while fee income from its key wealth growth pillar was steady but unspectacular. Revenue for HSBC's wealth division was flat at $2.7 billion. That compared with 15% and 23% year-on-year revenue growth in the first two quarters. The performance of HSBC's wealth business is under particular scrutiny because the British bank, like its peers, is focusing growth on fee-generating businesses such as wealth management as lending comes under pressure from falling interest rates.”