Banks grab AI-generated tiger by the tail

NEW YORK, June 26 (Reuters Breakingviews) - Artificial intelligence promises to help banks with two things they produce in abundance: profit and paperwork. The rapidly developing technology should help boost the first and reduce the second. For such a seismic and seemingly universal force for financial good, however, there’s little tangible sign yet that shareholders believe in the prospective value uplift for the industry.

More efficient lenders ought to be an easy sell to anyone who has endured excruciating loan application processes or been stuck at a store with a credit card mistakenly declined for suspected fraud. Generative AI, which produces human-friendly discourse from existing information, is on track to improve support for customers and pitch them suitable products. On the inside, it can do something similar, allowing employees to efficiently extract rapid and useful insights from mountains of unstructured data such as mortgage agreements and meeting notes. If that’s possible, as bank CEOs and software engineers generally agree, then financial institutions should theoretically be more attractive investments, too.

A $2 billion buyback is a welcome start, but the boss also needs to deliver on the promise of both an earlier restructuring and his own strategy reboot.

All the people and time spent churning out and sifting through documentation adds up to huge amounts of money. The bottom lines of banks worldwide could rise by 9%, or $170 billion, by 2028 thanks to generative AI, Citigroup analysts estimate based on a poll of financial-sector clients. A study from consulting firm McKinsey came up with a similar sum, between $200 billion and $340 billion, with the biggest impact in software engineering and customer-facing businesses.

Large numbers are only helpful up to a point, because of AI’s sprawling effects. JPMorgan boss Jamie Dimon said last month that AI could assist across his mega-bank, from travel plans to equity trading, and would basically “blow people’s minds.” His Morgan Stanley counterpart, Ted Pick, says tools such as a Zoom-embedded assistant that creates client meeting summaries for wealth managers could save them up to 15 hours a week, and are “game-changing.”

More tangible examples are also starting to emerge. JPMorgan has rolled out IndexGPT, a bot for big funds that uses keywords to suggest stocks to go with investment themes. Bank of America’s AI retail-bank assistant, Erica, is being introduced to business customers. Goldman Sachs is prioritizing code-writing, which consumes up to half the working day for a quarter of its staff.

Even a sophisticated chatbot would struggle to explain why bank investors are unmoved by the hype. The S&P 500 Index owes most of its 15% increase this year to the so-called Magnificent Seven group of AI-fueled stocks, led by chipmaker Nvidia. An exchange-traded fund that tracks such companies is up more than 30% compared to just 6% for the KBW Nasdaq Banks Index. The banking sector overall trades at just 0.9 times estimated year-ahead book value, according to LSEG data, hardly indicative of a mind-blowing transformation.

One reason is that AI is still mostly about potential. JPMorgan says it has 400 “use cases” already, a number that might double by the end of 2024, but there isn’t much shareholders can do with that information. It’s no coincidence that some of the biggest AI beneficiaries so far are consultants. Accenture has booked, opens new tab $2 billion of contracts over the past three fiscal quarters alone on fees related to generative AI.

While banks might in theory benefit from the technology more than many other non-financial companies, they also have a lot to lose. Mistakes can be dire where trillions of dollars are held. Citigroup’s fat-finger errors, including the accidental trigger of a $444 billion stock trade, exemplifies the value of AI that can spot slips and hints at the fallibility of the people who would be overseeing it. Algorithms might make mistakes, but regulators and litigious stakeholders will blame humans.

Investors, meanwhile, are zeroing in on potential savings. At Citi, boss Jane Fraser is making it her mission to “bend the expense curve.” For all the efficiency drives and technological advances, however, expenses at the biggest U.S. banks eat up 59% of revenue on a four-quarter average basis, the same proportion as in 2007, according to Federal Deposit Insurance Corp data.

Machine learning should help, but will hurt first. For each $1 spent on generative AI, there’s a further $3 deployed on “change management,” such as retraining staff or monitoring performance, McKinsey reckons. Dimon has warned that costs will go up, and that in some cases trying to measure the return will be a “waste of time.”

For now, costs are modest. Bank of America Chief Executive Brian Moynihan said last year that AI would probably account for roughly 15% of the $3.8 billion his lender spends each year on new technology initiatives, which is in turn around one-third of its total technology budget. Apply the same ratios to JPMorgan’s $17 billion tech allotment and it suggests the AI tab will be less than $1 billion.

Even when results do appear, there are good reasons for investors to be skeptical about how much they’ll benefit. Banks ought to be more predictable and dependable as they get better at, say, sidestepping fraud and avoiding regulatory penalties. Many of the AI-related dividends, however, will be passed onto customers, as big lenders compete to win business from each other.

Moreover, customers themselves are likely to get smarter. One example: Depositors rarely switch lenders, even when there’s a far better offer elsewhere, evidenced by the 17-year duration of the average U.S. retail-account relationship, according to personal finance website Bankrate. They might flee more readily if armed with real-time information, along with AI-powered instructions and processing capabilities. It would make bank funding models look very different.

The same could be true of credit cards, a business that’s lucrative partly because customers overestimate their ability to pay on time and overvalue opaque rewards. Wealth management also might be changed, as much of its value derives from information asymmetries between service providers and recipients. In a world of smart bots, the premia banks charge for such advice could shrink.

Years from now, AI will be everywhere in banking, and its value will be just as hard to quantify as other ubiquitous technology. Smartphones, electronic spreadsheets and online trading all have changed banking for the better. But with such opportunities comes understandable anxiety that capital will be misallocated, people will keep erring and gains will be competed away. The advent of generative AI might turn out to be banks grabbing a tiger by the tail.