At Least Three of the Five Largest US Banks Will Disclose a Quantified AI Savings Figure by End of 2027

The Prediction

By December 31, 2027, at least three of the five largest US banks by assets — drawn from JPMorgan Chase, Bank of America, Citigroup, Wells Fargo, and Goldman Sachs — will each disclose, in an official filing, earnings call, or named-executive statement, a specific quantified productivity or cost-savings figure explicitly attributed to generative or agentic AI (for example, "X percent engineering productivity," "$Y in operating savings," or "Z fewer support contacts handled by humans").

Confidence: 68% · Tier 2 (mid-term) · Target: December 31, 2027

Why I Believe This

The setup is already in place. JPMorgan has moved roughly $2 billion of annual AI spend out of its experimental innovation budget and into core infrastructure, alongside payment systems and data centers — the accounting move we analyze in How AI Became Infrastructure at JPMorgan. Once a cost is booked as infrastructure rather than as an experiment, the institution acquires a strong incentive to justify it publicly with numbers, and the bank has already floated one: an internal coding assistant credited with up to a 20 percent engineering productivity gain.

The competitive dynamics push the same direction. Bank earnings calls are imitative; once one CEO quantifies an AI return, the others face analyst pressure to show their own. Investor-relations teams increasingly want a clean AI number to point to, because "we spend billions on AI" invites the obvious follow-up of "and what did it return?" The path of least resistance is to publish a favorable, carefully-scoped figure.

What Holds Me Back From Higher

What keeps this at 68 percent rather than 85 is measurement honesty. Jamie Dimon himself has called technology returns elusive and hard to attribute, and self-reported AI productivity figures are notoriously soft. Banks may prefer vague language ("meaningful efficiencies," "significant productivity gains") precisely to avoid being held to a falsifiable number later. The prediction requires a specific figure, not a vibe — and the incentive to stay vague is real. That tension is exactly what makes this a clean target.

What Would Falsify This

• Fewer than three of the five named banks disclose a specific, AI-attributed quantified productivity or savings figure before December 31, 2027.
• Banks discuss AI spend heavily but only in qualitative terms, never attaching a concrete number to a return.
• AI spend is quietly folded into general technology results with no AI-specific attribution at all.

Signposts to Watch

• Quarterly earnings calls where a bank CEO or CFO volunteers an AI productivity or savings figure unprompted.
• 10-K / annual-report language moving from "we are investing in AI" toward attributed efficiency outcomes.
• Peer reclassification of AI into core-technology budget lines (the leading indicator that a disclosure number is coming).
• Headcount-per-output ratios shifting in support and engineering functions.

This is the falsifiable counterpart to the reclassification thesis: if AI has truly become infrastructure on the bank's books, the quantified number is a matter of when the disclosure pressure forces it out, not whether the figure exists internally.