The return of personalised banking

Banking has spent decades removing friction from the customer experience. The next opportunity may be to put something back: understanding.

There was a time when knowing your customer meant something quite literal. A local bank manager might know your circumstances, understand your business and discuss a lending decision in the context of a relationship developed over years. That model had considerable limitations, and there are few customers that would willingly exchange today's convenience for branch opening hours and paper-based processes. But as banking became more centralised, standardised and digital, a lot of context and understanding was lost in the process.

Sir Alan Budd, drawing on his experience as Group Economic Adviser to Barclays between 1988 and 1991, described an earlier banking environment where knowledge and long experience of customers' financial affairs were important to lending decisions. Acquiring that knowledge was time-consuming and costly, but could be regarded as an investment by both sides of the banking relationship.

A changing relationship between banks and their customers

Over subsequent decades, credit scoring, centralisation, telephone banking, internet banking and eventually smartphones changed how customers interacted with their banks. The result has been extraordinary convenience. We can open accounts from our phones, make payments instantly, freeze cards in seconds and access services at almost any hour.

But convenience and understanding are not the same thing. As we head towards 2027, advances in data, automation and artificial intelligence create the possibility of combining the two.

What if digital banking could remember?

Imagine a bank that does not simply store your information but can use it appropriately to understand your circumstances. It remembers information you’ve already provided, helps you understand where your money is going and recognises when something appears unusual. If you need help, you can describe the problem naturally rather than working out which department, menu or process owns it. Most importantly, you do not have to explain the same story repeatedly as you move between channels.

This is the banking experience I find most exciting, the opportunity to give both customers and the people serving them far more useful context. And elements of that future are already emerging.

From experimentation to production

One of the reasons I believe this conversation is becoming more important is that AI in financial services is no longer confined to experimentation. The Bank of England and FCA's 2026 survey gives us perhaps the clearest picture yet. Across all 196 respondents, 86% reported using AI, compared with 75% in the 2024 survey. The Bank of England cautions, however, that differences in survey population and scope mean the figures are not strictly like-for-like. Among the 105 firms completing the full 2026 survey, 94% had at least one AI system in production and generative AI accounted for 56% of the AI systems reported. Operations and IT accounted for around 31% of reported AI systems, while enterprise productivity, including applications such as coding assistants and summarisation, was the most widely reported current use case.

This data suggests an important transition. The first phase of enterprise AI has largely been about helping people work differently and the opportunity ahead is increasingly about helping organisations serve customers differently.

The front door is becoming conversational

Some of the clearest signs are appearing in ordinary banking moments. Whether that’s understanding a bill, questioning a payment, exploring spending or getting help when something goes wrong. Here are a few examples of how banks are putting this into practice…

Revolut: conversation connected to action

Revolut is using ElevenLabs Agents for voice support across the UK and Europe, connecting the experience to its proprietary banking systems through secure interfaces. ElevenLabs provides the conversational layer, while Revolut retains control over orchestration and business logic. During the first rollout phase, ElevenLabs reported that time-to-resolution improved by more than eight times. The important point is not simply that customers can talk to AI. The true value comes from connecting the conversation to the systems, information and processes required to resolve the customer’s problem.

Starling Bank: letting customers interrogate their money

Starling Bank's move to Google BigQuery created a foundation for analysing its data, while Vertex AI supports AI development and evaluation. Gemini can interpret customers' questions about their spending and trigger Starling's analytics capabilities to calculate an answer. The first version of its Scam Intelligence capability was developed in four weeks.

The conversational experience did not appear independently of the bank’s data environment. AI did not remove the need to fix the foundations either, it simply increased the value of doing so.

NatWest: personalisation means understanding how customers speak

NatWest's Spending Insights combines NVIDIA Nemotron automatic speech recognition with NatWest's proprietary small language models. Initial testing involved 90 colleagues and around 3,000 banking questions. The results showed a 20% relative reduction in words misheard across the accents tested. The published material does not provide the baseline word-error rate, so this should not be interpreted as a 20-percentage-point increase in overall accuracy.

In financial services, confusing a merchant, amount, date or phrase can fundamentally change the question being asked so accuracy becomes part of the customer experience.

Monzo: adding context to fraud detection

Monzo's Fincrime Agent platform uses language models to process unstructured customer context alongside predictive rules and supervised machine learning. Human investigators still remain involved I instances where definitive automated resolution is unavailable.

The lesson here is not that generative AI replaces established AI. Language models help understand context, predictive models identify patterns, automation connects processes and people intervene where judgement or accountability requires it.

Does the less visible transformation matter more?

For customers to experience a bank as one joined-up organisation, it must behave like one behind the scenes. That is difficult when customer information, policies, workflows and decision-making processes are distributed across decades of technology.

Lloyds Banking Group: hiding organisational complexity

Across more than 10,600 live customer cases over 90 days, Lloyds Banking Group reported that the number of systems fraud specialists needed during a customer call had been roughly halved from about 12 to 15 previously. Trained fraud specialists retained accountability for customer decisions. What this tells us is that a good banking experience should not require customers to understand the organisational or technological structure of their bank in order to get something done.

Nationwide: personalisation starts with the data estate

Nationwide consolidated 19 separate data stores onto Microsoft Azure and Azure Databricks, creating a more unified data platform. It has also used GPT-4 through Azure OpenAI to support customer correspondence, with human review retained. The reported time required for this work fell from around 45 minutes to 10 to 15 minutes. Approximately 2,500 engineers were reported to be using GitHub Copilot.

Although it is tempting to focus on GPT-4, I think the more important part is the 19 data stores. Personalisation depends on having the right information available, connected and governed at the moment it is needed.