Bank of America is about to spend a lot more on AI, and the numbers behind that decision tell you more about the state of corporate AI than any product launch could.
The bank's Co-President Jim DeMare and CEO Brian Moynihan laid out the math at recent investor conferences. About 140 AI projects have been built so far, costing 400 million dollars combined and generating 800 million dollars in measurable benefit, a two-to-one return. Based on that, the bank plans to double its AI spending next year.
Its chief technology officer added one more detail: internal demand for AI projects is already running twice as high as last year, driven by employees pitching their own ideas for what AI could fix in their jobs.
Here is the part worth paying attention to: the wins that are actually provable are narrow. Coding tools are speeding up software development by 15 to 20 percent, a gain showing up across most industries, not just banking. And an internal assistant called Erica is now handling enough routine employee questions to do the work of roughly 11,000 help desk staff.
Both of those are backend, technical tasks with clear before-and-after numbers. Nobody at the bank has shown a client-facing AI win with the same clarity yet, which tells you where the technology is actually mature enough to trust today.
Meanwhile, the bank's headcount slipped from 213,000 to 209,000 employees this year. Management frames this as attrition, not layoffs, meaning they are simply hiring more slowly and letting an 8.5 percent turnover rate do the shrinking.
That distinction matters less than it sounds, because the effect on the job market is the same either way: fewer new positions open up. Bank of America is not alone in this. The six largest US banks together cut 15,000 jobs in the first quarter of 2026 while posting a combined 47 billion dollars in profit, and their executives are now willing to say AI played a role, something banks avoided admitting openly just a year earlier.
The uncomfortable number in all of this comes from Accenture, which surveyed hundreds of bank executives and found that only 20 percent of them are seeing AI value that is both widespread and lasting across their organization. Most of the money going into AI right now is still funding experiments, not proven results.
Bank of America's own leaders openly say fear of falling behind competitors is part of what is driving the spending, not certainty about returns. That is worth remembering the next time a spending number gets reported without context.
For any business watching this from outside banking, the lesson is not to copy the spending, it is to copy the discipline. Bank of America can point to exact costs and exact returns for its narrowest, most measurable AI projects, but it cannot yet do the same for its bigger, vaguer promises.
Any AI investment your business makes should start where you can measure the return in a specific number, the same way Bank of America did with its coding tools and its help desk, before chasing the parts of AI that only sound impressive.