AI agents could cost banks $500bn — by winning savers better rates

It is a fact of banking that the less you care, the more your bank earns. Lenders make a handsome profit by taking money customers leave idle in their bank accounts and investing it at higher rates. Agentic AI, the buzz-phrase of the moment, will upset this cosy arrangement. And while it would take more than that to knock banks off their perch, they have quite a lot to lose.

The theory is that AI assistants like Meta’s Muse could help customers put trillions of dollars of “lazy money” to better use. That could just mean flagging when a saver could get a better deal on their surplus balance. Or more ambitiously, automatically sweeping the excess into a money-market fund offering more than the nearly nil paid on instant-access deposits today.

This would be a big change. Banks make about two-thirds of their income on the spread between what they pay and what they lend. JPMorgan Chase, Bank of America and Wells Fargo hold $1.6tn of non-interest bearing deposits — 16 per cent of their liabilities. Paying a 3 per cent rate on those would cost them $47bn a year, almost half of their collective annual earnings.

Will bank profit really evaporate to that extent? Not likely. Shrill talk of “agentic bank runs” reckons without the manifold frictions that stop customer funds from flying hither and thither. The simplest barrier is trust: will households hand control over their finances to companies known for moving fast and breaking things? Who takes the rap if the agent sends funds to the wrong place?

But AI will no doubt make customers less inert — and there’s a way to estimate what that will cost their banks. Whenever a lender buys another in the US, it must quantify the financial benefit from acquiring sticky, cheap deposits, as opposed to raising funds in the market. This is called a “core deposit intangible”, and is , according to analysis by Mercer Capital.

Run that back-of-the-envelope ratio across all US banks, with their $20tn of deposits of which roughly 80 per cent are “core”, and it comes out at about $500bn. That, in other words, is the chunk of equity value banks would lose if all of those deposits received market-like returns. It’s about one-tenth of the total value of listed US banks, based on S&P Capital IQ data.

True, banks will gain from AI as well as lose. Cutting costs, pricing lending more effectively and avoiding costly human error — no more Citigroup-style bungled $900mn wire transfers — should all generate gains. Bank of America chief Brian Moynihan said last month that his institution spent $400mn on AI improvements and saw $800mn of benefits. As agents gain traction, the fruits of efficiency drives will increasingly flow to depositors as well as shareholders.

For now, there are bigger things to fret about. More than half of bank investors think Federal Reserve rate hikes are the biggest risk to valuations, lender Truist found in an investor survey at the end of September. Very few cited the threat from bots. And as JPMorgan chief Jamie Dimon regularly warns, the banking system hasn’t weathered a proper credit downturn for well over a decade, and one is long overdue. The real agents of chaos are already with us.

john.foley@ft.com