On September 14, at a Barclays conference, Bank of America CEO Brian Moynihan put a number on the bank’s AI program: 130 to 140 projects in place “at a cost of $400 million, generating a benefit of $800 million.” The bank has never given a dollar figure like that on an earnings call. On October 14 it reports third-quarter results, its first call since Moynihan spoke. Will the $800 million make it onto the call?
I wouldn’t bet on it. My team went through 919 earnings calls of 60 major U.S.-listed financial firms since January 2023. One firm put a specific dollar figure on what AI had saved or earned it: S&P Global, about $9 million a year from its Kensho Scribe tool. Analysts asked for such a number 29 times and never got one.
Among 100 major U.S. companies outside tech, communications and finance, only FedEx has done so on a call, crediting “a new AI image capture process” with a surcharge benefit of “over $180 million annualized.”
That makes two companies in 160.
“We don’t have the full confidence to put information out publicly,” Goldman Sachs CEO David Solomon told the analyst Mike Mayo in January.
In June 2025, at a Morgan Stanley conference, JPMorgan Chase CEO Jamie Dimon said, “we do know we get benefits around $2 billion to pay for the $2 billion.” JPMorgan has not said that figure on any of its 13 earnings calls since mid-2023, and in October 2025 its CFO, Jeremy Barnum, told analysts that proving AI savings “turns out to be a very hard thing to do, hard to prove.”
None of this makes me skeptical about the eventual gains from AI investment for financial institutions and across the economy. AI is an all-transforming technology that will benefit most organizations at some point.
There are three related matters that do make me skeptical, though.
The first is the calendar. AI investment will pay off on what economists call the productivity J-curve. A bank that cannot yet put an AI number on its call is where the curve says it should be. This rhymes with history. A general-purpose technology first needs investment in things that national accounts barely measure, such as new processes and skills. In 1987 Robert Solow wrote, “You can see the computer age everywhere but in the productivity statistics.” A 1994 survey of 99 completed reengineering projects found 67% had produced “mediocre, marginal, or failed results.” Yet by the end of the decade information technology accounted for about two-thirds of the pickup in U.S. productivity growth.
Second is the fact that even real, materialized AI edge fades fast. China’s long-only quant funds, whose boom Bloomberg tied to their rapid adoption of AI, beat human stock pickers by 20 percentage points in 2025, saw their edge over the index halve by April, and then lost an average 17% in July.
Finally, there’s the cost. Inside the firms, the talk has already firmly turned to this concern. AI cost came up on 53% of the financial calls we read last season, up from 3% in late 2022. The 40 engagements I have led with financial firms since 2025 suggest that optimization can remove about 40% of annual AI spend. The most intense users are cutting their token bills as they optimize for a return metric, becoming supplier-agnostic in the process. The age of tokenmaxxing, burning tokens to show adoption, is already over.
Put all this together and you have a technology whose users are already cost-sensitive, one that will take time to show value, only for that value to accrue to a small sub-segment of adopters and then erode quickly. This is far from the picture of near-certain skyrocketing of demand that is fueling big lab valuations and soon the IPO materials.
The incentives to talk up one’s own AI use are clear: anyone raising money now needs to look like an AI firm, one of the emerging winners of the revolution.
But allocators should pause. Right on the trend and wrong on the calendar is a way to lose money as old as investing itself. Economists such as Carlota Perez have described the workings of technological bubbles in detail. The most remarkable thing about the AI cycle might be how closely it follows the blueprint.
So, here is what to listen for next week. JPMorgan, Goldman Sachs, Wells Fargo and Citigroup report on Tuesday, Oct. 13, and Bank of America and Morgan Stanley on Wednesday. Listen for a specific dollar figure for what AI saved or earned, said on the call itself. If an analyst asks, listen for whether the answer is a percentage, a headcount or “hard to prove.” If Moynihan repeats his $800 million, Bank of America will be the first of the 21 banks we track to put such a number on a call.
Milos Maricic is the founder of SPEC Research, which helps large allocators select AI and quant managers, and of Maximand, which helps financial institutions optimize AI token ROI.
The opinions expressed in Fortune.com commentary pieces are solely the views of their authors and do not necessarily reflect the opinions and beliefs of Fortune.