Artificial intelligence undoubtedly offers enormous potential to banks. Entities ranging from Bank of America to Société Générale are already using large language models to cut costs and increase productivity. But AI could also help customers. If account holders and borrowers use intelligent “agents” to automatically search for the best deals, deposits could become more volatile and credit spreads narrow. So the real question is how much of the benefits of AI will banks be able to retain over time.
Chatbots like Claude and ChatGPT can now shorten customer service calls, reduce IT help desk queries, and dramatically lower software development costs. Increasingly, AI will make other important and costly back-office tasks more efficient. For example, anti-money laundering and fraud controls, but also non-regulatory tasks such as credit analysis. AI could probably analyze, bundle and securitize loans more quickly.
The financial benefits of all this are beginning to emerge. Bank of America CEO Brian Moynihan has boasted about $800 million annually in cost cuts and revenue increases through the use of generative AI, compared to a one-time initial outlay of just $400 million. That points to an impressive return on investment. In Europe, meanwhile, banks are beginning to incorporate AI into their financial objectives. Société Générale has recently identified up to €600 million in savings thanks to AI, almost a third of all cost cuts planned within a new strategy. Commerzbank has said it expects annual profits of 500 million from AI from 2030.
Profits will not appear overnight. Banks have to consolidate myriad IT systems and move data from on-premises servers to the cloud. Still, there are reasons to think that long-term savings could exceed current early estimates. An executive at a European bank, for example, told Breakingviews that he expects AI to reduce workforce by 20% within five years. Applying the general rule that personnel expenses account for 60% of the total, that alone would imply a reduction of at least 12% in the group's expenses.
UBS analysts, for their part, have crunched numbers with the Nordic banks, considered among the most advanced in digital. They estimated a 15%-18% gross reduction from current cash expenses. The only catch, they say, is that almost half of the savings could be absorbed by the additional spending required on technology suppliers and systems, leaving a net efficiency improvement of perhaps 9%.
Who keeps the savings
But it remains to be seen whether it will be shareholders who benefit from these efficiencies. Banks operating in competitive markets, such as British mortgages or US wealth management, will be under pressure to pass on the savings generated by AI to customers through lower fees and cheaper loans. An even bigger problem is that AI can help individual customers and businesses compare offers much more easily.
The main front of attack is the deposits. Banks still make a lot simply by taking in cash, often in barely paying checking accounts, and investing it in securities or depositing it with the central bank. AI could change that. Clients could ask an agent, like Muse, Meta's personal assistant, to determine how much liquidity they really need and move the rest into high-paying savings accounts, money market funds and similar products. It would be as if each person had their own company treasurer, charged with squeezing every last cent out of excess liquidity and ensuring that banks don't keep the profits from idle money.
The figures are scary. Banks covered by the US Federal Deposit Insurance Corporation pay an average of 0.1% for checking accounts and 0.4% for savings accounts, compared to between 3% and 5% offered by a range of financial technology companies. This is indicated by figures compiled by Apollo chief economist Torsten Slok. The risk for banks is that AI agents cause an avalanche of money leaving traditional entities in search of higher returns. Something similar could happen in Europe. The 20 largest listed banks based in the region – including HSBC, UBS and Intesa Sanpaolo – generated around €326 billion in net interest income in 2025, around 60% of their total income.
The threats multiply. Muse appears to have caused concern in the prices of large banks such as JP Morgan and Wells Fargo after its launch. And there are many smaller fintechs preparing too. One of them is Riff, in the United Kingdom, which plans to soon launch a savings agent. Its founder, Oliver Butcher, estimates that money deposited in British current accounts could fall by around 80% in a scenario in which AI agents optimized customers' savings after setting aside enough to cover 40 days of expenses.
Banks have reason to think they will avoid the AI-pocalypse. Many customers will be wary of putting control in the hands of a robot. And those with small deposits will probably be less willing to actively seek higher compensation. Still, it is difficult to imagine that the banks will emerge unscathed. As AI agents become more widespread, entities will be pressured to offer their own versions of these products to maintain customer relationships. Interest margins seem likely to contract, having grown in Europe by a third from the 2021 low, to an average of 1.63%. In the US, Bank of America analysts estimate that just over 20% of deposits do not earn interest. This is much lower than the peak of over 30%, but still higher than the pre-2008 level.
Other lines of business, such as wealth management and insurance, will also come under pressure. These fee income accounted for almost a third of European banks' total operating income in the second quarter, according to the European Banking Authority. The emergence of AI tools will make financial advice in areas such as portfolio allocation and tax management much cheaper and more competitive. More sophisticated clients may still be willing to pay for a tailored service, but banks ranging from UBS to Morgan Stanley and Goldman Sachs will compete for a shrinking segment of high-margin business.
It is striking that, except for some recent hesitation, shareholders are not worried. The large US banks are trading at 12 times expected profits, and the European ones at 10. Both figures exceed their respective averages of the last 15 years, of 11 and 9 times. In other words, investors do not foresee a collapse in profits. But neither do these valuations reflect a huge windfall thanks to AI. The bottom line is that if the benefits of large language models are realized, banks may have a hard time keeping them.
