Banks will soon face the dark side of AI

Intesa Sanpaolo

30/09/2026 17:50

The author is a Reuters Breakingviews columnist. The opinions expressed are his own.
By Neil Unmack
LONDON, Oct 1 (Reuters Breakingviews) - There's little doubt that artificial intelligence holds huge promise for lenders. Players from Bank of America BAC.N to Société Générale SOGN.PA already use large language models to cut costs and boost productivity. Yet AI could help customers too. If account holders and borrowers use smart "agents" to automatically find the best deals, then deposits could become flightier and lending margins tighter. So the real question is how much of their AI gains the banks will be able to hold onto over time.
Chatbots like Anthropic's Claude and OpenAI's ChatGPT can already shorten customer service call times, reduce IT helpdesk queries and slash software-development costs. Increasingly, AI will make other large and costly back-office jobs more efficient. Examples include anti-money laundering and fraud checks, but also non-regulatory tasks like credit analysis. AI could probably analyse, parcel up and securitise loans more quickly.
The financial benefits of all this are starting to emerge. BofA CEO Brian Moynihan has touted some $800 million of annual cost cuts and revenue gains from using generative AI, from just a one-off upfront cost of $400 million. That suggests an impressive return on investment. In Europe, meanwhile, lenders are starting to hardwire AI into their financial targets. France's SocGen recently identified up to €600 million of AI savings, accounting for nearly a third of its total planned cost cuts as part of a new strategy. Germany's Commerzbank CBKG.DE has said it expects annual AI benefits of €500 million by 2030.
The gains won’t appear overnight. Lenders need to consolidate myriad IT systems and move data from local servers to the cloud. Even so, there is reason to think that the long-term savings could eclipse today's early estimates. One European bank executive, for example, told Breakingviews that they expect AI to enable a 20% reduction in headcount after five years. Using a rule of thumb under which staff costs make up 60% of the total, this alone would imply at least a 12% overall reduction in group expenses.
UBS analysts, meanwhile, crunched the numbers on Nordic banks, which are deemed to be among the most digitally advanced. The researchers estimated a 15%-18% gross reduction in current cash expenses. The only wrinkle is that nearly half of those savings, the UBS analysts reckon, could be eaten up by necessary extra spending on technology providers and systems, implying a net efficiency gain of perhaps 9%.
Yet it remains to be seen whether these efficiencies will accrue to shareholders. Banks in competitive markets, like UK mortgages or American wealth management, will be under pressure to pass the boost from AI savings on to customers through lower fees and cheaper loans. An even bigger issue is that AI tools may help retail and corporate customers shop around more seamlessly for a better deal.
The biggest area of attack is deposits. Banks still make a lot of money simply by taking in cash, often in the form of low-yield checking accounts, and investing it in securities or stashing it at the central bank. AI could change that. Customers could ask an agent, like Meta Platforms' personal assistant Muse, to work out how much liquidity they really need and shift everything else to high-yield savings accounts, money-market funds and the like. It would be like every individual having their own corporate treasurer, whose job involves wringing every possible cent out of spare liquidity and making sure that banks aren't keeping the benefits of idle cash for themselves.
The numbers look scary. Banks covered by the US Federal Deposit Insurance Corporation on average pay 0.1% on checking accounts and 0.4% on savings accounts, versus the 3% to 5% on offer at an array of financial technology upstarts. That's according to figures compiled by Apollo Global Management Chief Economist Torsten Slok. The risk for banks is that AI agents cause a flood of cash to leave mainstream lenders in search of higher returns. Something similar could happen in Europe. The top 20 listed lenders based in that region — including HSBC HSBA.L, UBS and Intesa Sanpaolo ISP.MI — cranked out some €326 billion in net interest income in 2025, accounting for around 60% of total revenue. 
The threats are proliferating. Meta’s Muse seems to have caused jitters in the share prices of big banks like JPMorgan JPM.N and Wells Fargo WFC.N after its recent launch. There are many other smaller fintechs lining up too. One is Riff, in the UK, which is soon planning to launch a savings agent. Its founder, Oliver Butcher, estimates that the amount of money sitting in UK current accounts could fall by around 80% in a scenario where AI agents optimise customers' savings after setting aside enough to cover 40 days of expenses.
Banks have reasons to think that they will avoid an AI-mageddon. Many customers will be wary of handing control to a robot. And those with small deposits are less likely to actively chase higher rates. Still, it's hard to see banks escaping unscathed. As AI agents spread, lenders will face pressure to offer their own versions of these products to keep hold of the customer relationship. It seems likely that net interest margins will contract, having risen by roughly a third in Europe since the 2021 trough to an average of 1.63%. In the US, BofA analysts estimate that just over 20% of deposits bear no interest. That's down sharply from a peak of over 30%, but is still above the pre-2008 level.
Other business lines, like wealth management and insurance, will feel the heat too. Such fee-based revenue accounted for nearly a third of the total operating income across European banks in the second quarter of 2026, per European Banking Authority data. The emergence of AI tools will make financial advice like portfolio allocation and tax management much cheaper and more competitive. Sophisticated clients may still pay up for a bespoke service, but banks from UBS UBSG.S to Morgan Stanley MS.N and Goldman Sachs GS.N will all find themselves chasing this dwindling pool of high-margin business.
It's notable that, a few recent wobbles aside, shareholders aren’t fretting. Big US banks currently trade at 12 times forward earnings, while the Europeans are at 10. Both numbers are above their respective 15-year average multiples of 11 and 9. In other words, investors do not see a profit collapse coming. Nor, however, do these valuations imply a huge AI windfall. The implication is that if the benefits of large language models do arise, banks may struggle to hold onto them.
     
Follow @Unmack1 on X.

(Editing by Liam Proud; Production by Maya Nandhini)
((For previous columns by the author, Reuters customers can click on UNMACK/neil.unmack@thomsonreuters.com))