Banks are the plumbing … we just hope they don’t leak

The future of banking is being fought over by artificial intelligence, human intelligence and, occasionally, a complete lack of either.

Two headlines caught my eye yesterday, and they made me laugh. Not because they were particularly funny, but because they captured the glorious absurdity of where banking is heading.

The first, from FinTech Global, asked whether banks can explain every action taken by an AI agent to the Financial Conduct Authority (FCA). The second, from Banking Dive, argued that banks should emphasise the human touch to differentiate themselves in the age of artificial intelligence.

So, on the one hand, we have banks rushing to replace humans with robots and, on the other, we have banks being told that their greatest competitive advantage is being human.

You couldn’t make it up.

We have spent the last twenty years telling customers to stop bothering us and to do everything yourself. Now we are discovering that customers don’t particularly like talking to machines and would like a human engagement every now and again, particularly when something goes wrong.

Who knew?

This contradiction goes straight to the heart of something I discussed in my keynote at Visa Next earlier this week. Banking is being hacked, automated, reinvented and disrupted all at once, but the fundamental principles of finance haven’t changed.

Trust allows commerce to happen. Transactions make commerce work. Relationships make commerce sustainable.

And we are in danger of forgetting the third whilst obsessing about automating the second.

Let’s begin with the regulatory problem.

Imagine that you give your personal AI agent permission to manage your finances. It knows your income, expenditure, savings, investments, mortgage and credit cards. It understands your financial circumstances better than your bank, your accountant and, quite possibly, your spouse.

You tell it to optimise your money.

Within milliseconds, it discovers that your bank is paying you a miserable rate of interest on your savings whilst offering new customers something far more attractive. It negotiates with the bank’s AI agent, gets nowhere, finds a better deal elsewhere and moves your money.

Wonderful.

Except your bank notices a robot accessing your account and panics.

“Security alert! Unauthorised artificial intelligence detected!”

Your agent replies, “I’m authorised by the customer.”

The bank responds, “How do we know?”

Your agent says, “Because the customer told me.”

The bank says, “We don’t recognise you.”

Your agent replies, “That’s funny. My customer doesn’t recognise half the charges on their account either.”

Meanwhile, the FCA wants to know who authorised the transaction, how the decision was made, whether the customer understood the consequences and who is accountable if something goes wrong.

The bank explains that its AI system made the decision.

The regulator asks why.

The bank consults explains that it made the decision because its algorithm determined that the decision was appropriate.

The regulator asks what that means.

The AI system replies, “Would you like me to connect you to a human?”

Unfortunately, the bank made the human redundant last Tuesday.

Welcome to the future of finance.

This isn’t entirely fantasy.

American banks have already been restricting third-party AI agents from accessing customer accounts because they cannot always distinguish legitimate customer instructions from unauthorised activity, and there are genuine security issues here. An AI agent that can access accounts, initiate payments and move money creates risks that banks cannot ignore. Permission to check a balance is not permission to empty an account, and a customer who authorises one transaction has not authorised a thousand others.

The FCA is examining the implications of agentic finance including identity, delegated authority, accountability and execution. The important point is that the regulator isn’t demanding an explanation of every mathematical operation inside a neural network. It wants financial firms to demonstrate that decisions are controlled, permissions are valid and responsibility is clear.

That is entirely reasonable.

What intrigues me is that banks are enthusiastic about artificial intelligence when it works for them but become distinctly nervous when it works for us, and this is where the real battle begins.

For decades, banks have enjoyed a significant information advantage over customers. They know the products, the margins, the fees, the risks and the alternatives. Customers, meanwhile, have been expected to navigate complex terms and conditions, compare incomprehensible offers and make sensible financial decisions whilst trying to remember seventeen different passwords.

Artificial intelligence changes that relationship.

My AI agent doesn’t get tired of comparing interest rates. It doesn’t forget to switch providers. It doesn’t get confused by introductory offers that expire after six months. It doesn’t accept a lousy deal because the paperwork is too complicated.

More importantly, it doesn’t have an emotional attachment to my bank’s logo.

If another bank offers a better deal, it moves my money.

Banks have spent billions creating customer loyalty programmes, building brands, developing apps and analysing behaviour. Now they face a technology that could make customer loyalty disappear in milliseconds.

That’s rather inconvenient.

Imagine a bank spending £100 million developing a beautiful new mobile banking application, only to discover that customers never open it because their AI agents do everything for them. The bank becomes an invisible utility, and the customer doesn’t care whether the money sits with Barclays, HSBC, Santander or the Bank of the Outer Hebrides, provided it is safe, accessible and earning a decent return.

Banks are just financial plumbing in the same way as supplying water or electricity. Very sophisticated financial plumbing, admittedly, but plumbing nevertheless.

Further reading

• Can banks explain every AI agent action to the FCA? — FinTech Global
• Banks should play up human touch to stand out in AI era — Banking Dive
• Banking is being hacked, automated, reinvented and disrupted … ALL AT ONCE — The Finanser

Postscript:

In 2025, English water companies discharged untreated sewage 289,838 times, lasting for more than 1.8 million hours, according to the Surfers Against Sewage Water Quality Report