For decades, the most consequential conversation in a customer’s financial life happened in a cubicle at the back of a branch: impersonal, slightly airless, an adviser clacking away on elevated keys.
Most of those branches have closed. The conversation has moved to a chat window, minus the banks.
Banks’ reflexive retort is to scold, chastise and discourage this behaviour. In my adopted homeland of Britain, this is practically a national pastime.
It’s not working. Asking for financial guidance is now the second most common use of generative AI platforms, just behind health and wellness. Fifty-one percent of consumers already turn to AI for financial advice or information, rising to 82% among millennials and Gen Z. Year over year, the share of people using these tools for financial decisions more than doubled between 2024 and 2025.
The chiding has not been confined to bank branches. Futurism, surveying the landscape of AI financial tools, concluded that the models were “strikingly bad at giving financial advice.” The FCA flagged in a recent perimeter report the rapid growth of general-purpose AI tools offering financial advice or recommendations – framing it, for now, as a risk to be monitored rather than a resource to be mobilised.
The message from every direction has been consistent: go back inside, leave this to the professionals. The professionals are not exactly waiting.
In my anecdotal experience, this is how it actually happens. Gemini, already supplanting trip planning and cooking, becomes the default for simple, fact-based inquiries. What’s the current Stocks and Shares ISA limit? How much can you put in a Roth IRA? Increasingly, AI results are so prominent in search that no one needs to load an application at all.
From there, it’s natural to start to volunteer “hypothetical” information – last year’s salary, or the price of your home. The answers get more specific.
Then comes the Rubicon moment: that first payslip screenshot or tax return. Suddenly, there’s a hyper-smart computer in possession of all your personal information. And the advice gets more personalised.
Now the chorus of a thousand compliance officers: why would you feed your most personal data to a plagiarism machine? What in the track records of Musk or Altman would lead anyone to trust them to keep data private?
Like so many things, it is a hypothetical risk set against a concrete benefit. Maybe some information will get out — this seems to happen regardless of information hygiene. I once bought a brand new Patagonia jacket with the proceeds of one of those years-later class-action settlement notices that come in the mail.
Often, there is no practical alternative. Just 9% of UK adults received financial advice about their pensions or investments in the previous twelve months, according to the FCA’s Financial Lives survey. The FCA estimates around 23 million adults are underserved by the advice and guidance market — including 7 million with at least £10,000 in investable assets who could benefit from advice but aren’t getting it. There are financial advisers, but they are expensive and increasingly constrained by regulation — half of advice firms have now stopped serving certain clients as a direct result of regulatory obligations. There are robo-advisers, which also cost money and have struggled to prove their value.
The bar AI has to clear is comically low. There’s no need to wait for a consequential moment — the big promotion, or the first pregnancy. The machine keeps no hours and places no minimum on the size of problem it will engage with. A quarter of Americans say they avoid seeking financial advice because they’re insecure about their financial position — but 40% say they would have done so if the other end of the line had been a chatbot.
Increased access means the ability to dream. I found myself in a café, passing the time before a black tea arrived, running through what it would look like to spend 90% of my income on housing, or to max my salary sacrifice contributions. None of this was under serious consideration. But by playing through real scenarios, I was able to narrow the Overton window of actual action.
This phantom competitor cannot be willed away. It seems doubtful that the government will lock out an entire subject from permitted conversations. If the Monolith cannot be slain, traditional banks and challenger banks need a proportionally attractive response, one that is equal parts product and message.
Go wide, unfettered by the scope of the product road map. Customers still consider bond investments even if they aren’t on a particular investment platform; any bot that excludes that conversation becomes less useful.
One area of particular promise is the overall net worth dashboard — knowing, at a glance, what you own and owe across accounts, property, pension, and debt. Tools like Mint made a credible attempt at this in the 2000s before being acquired, neglected, and eventually shut down by Intuit in 2024. AI provides a genuine opportunity to revive and substantially improve a very sensible idea. Monument, a British mass affluent challenger bank and a client, now offers this to members. It is an excellent instinct.
Plaid’s partnership with ChatGPT, which allows 12,000 financial institutions to connect bank accounts, represents one path forward, but only if the bank integration is a meaningful upgrade over standard conversations.
Jealousy is a better motivator than guilt. The instinct to show customers what they’re missing – rather than lecturing them on what they’re doing wrong – is the smarter play. Yes, the real value of a bank’s AI offering is that it operates inside a regulated environment, with meaningful controls against hallucination and against the model deciding it’s time to go full tilt on meme stocks. But lead with the show, not the disclaimer.
Create that jealousy through specificity and example. Show full conversations that real customers have had inside the environment. Some discretion will be needed, but to the extent possible, let people see what is actually possible. A customer asks about pension contributions at 11pm; the model cross-references their salary sacrifice ceiling and finds an extra £3,200. Part of what makes these models both exciting and unsettling is that their potential range of functions is vast — but individual users are limited by their imagination. Detail expands imagination.
Done correctly, a bank maintains attention in a world where that is vanishingly thin. The right bank could find itself having built a true replacement for the branch, not a functional substitute but an entrepôt: a place where customers actually want to linger and think through where they want to go.
Jon Schubin runs content for Cognito
