Global banks are pouring billions into artificial intelligence, yet the payoff keeps feeling smaller than the press release. Leaders now admit the truth out loud. Most AI projects are not hitting the returns boardrooms expected, and analysts say the gap will not close until banks stop bolting AI onto old workflows and start rebuilding the work itself.
The Return Gap That Wall Street Cannot Ignore
The conversation around AI returns in banking has shifted from hype to hard math. Wells Fargo analyst Mike Mayo summed it up bluntly when he called AI in banking “not a silver bullet” and described the shift as a long, expensive and risk constrained transformation.
Mayo also pointed out that almost no bank, apart from JPMorgan, openly shares its AI savings. That silence speaks loudly. If the numbers were strong, more CFOs would be reading them out on earnings calls.
JPMorgan Chase remains the rare exception. CEO Jamie Dimon has confirmed the bank earns roughly $2 billion in annual AI generated savings, which is close to what it spends on AI each year. Break even at scale is progress, but it is not the windfall many predicted.
Why The Steam Engine Story Still Haunts Banking
Accenture’s global banking lead Michael Abbott has compared today’s AI rush to the early days of the steam engine. Mills first dropped a steam engine where the water wheel used to sit and kept everything else the same. The result was small productivity gains at a much higher cost.
Banks are repeating that mistake. They are sprinkling AI on top of legacy call centers, paper heavy onboarding and clunky credit workflows, then wondering why returns look thin.
The lesson from history is sharp. Real value only arrived when factories redesigned the entire process around the new engine. AI demands the same courage from banks today.
Three Mindset Shifts Separating The Winners
Banks that are pulling ahead share a common trait. They have moved past pilot projects and started rewiring how work gets done.
- From task to orchestration: Single use chatbots are out. Multi agent systems that hand off work across compliance, fraud and service teams are in.
- From productivity to growth: Top banks now ask whether AI is creating revenue, not just trimming headcount.
- From projects to platforms: Winners build shared AI foundations with proprietary data, instead of dozens of disconnected experiments.
Bank of America’s virtual assistant Erica now does the work of roughly 11,000 employees. The bank also confirmed that all 18,000 of its software developers use coding agents, with productivity gains close to 20%.
Goldman Sachs CEO David Solomon recently told shareholders the firm is building what he called an “AI propelled operating model”. Goldman is rebuilding processes first, then layering AI on top, the opposite of how most pilots begin.
The Numbers Behind The New AI Banking Race
Spending is accelerating even as ROI questions linger. Industry trackers report that bank AI budgets jumped by double and triple digit percentages over the last year, the steepest climb on record.
| Bank | AI Move | Reported Impact |
|---|---|---|
| JPMorgan Chase | Proxy IQ platform replacing outside advisors | Covers 3,000+ annual company votes |
| Bank of America | Erica plus developer coding agents | 11,000 employee equivalent, 20% dev gains |
| Lloyds Banking Group | Agentic GeoAI system | 800x faster search, 90% less energy use |
| BNY | Multi agent payment validation tools | Faster remediation across operations |
| CIBC | Microsoft Azure and Copilot rollout | Enterprise wide productivity push |
One striking shift came at JPMorgan, where the bank built its own AI platform called Proxy IQ and walked away from outside proxy advisory firms entirely. Dimon publicly called the old model broken.
Meanwhile Danske Bank named Fiona Browne as its first Head of AI, and State Street appointed Mark Wightman to lead transformation. These hires signal that AI is no longer a side project tucked inside IT.
Quick Take: 78% of banks investing in AI report a positive ROI within 18 months, yet only one major US bank publicly shares its dollar figure. The gap between belief and disclosure is now the story.
What Bank Leaders Should Do Right Now
The roadmap is becoming clear for boards that want AI to actually pay back. The winners are not the banks with the biggest models. They are the ones with the cleanest data and the boldest process redesign.
- Pick one core journey, like onboarding or dispute handling, and rebuild it end to end.
- Tie every AI use case to a revenue or trust metric, not just cost cuts.
- Concentrate spending on a shared platform instead of 50 small pilots.
- Be honest with investors about savings, even if the figure is small today.
Mayo flagged that roughly one third of bank jobs, or parts of them, may eventually shift to AI handling. That number will scare staff and excite shareholders in equal measure.
The banks that win the next decade will not be the ones spending the most on AI. They will be the ones brave enough to redesign the work around it.
The AI moment in banking feels a lot like that early steam engine factory floor. The technology is here, the noise is loud and the smoke is everywhere, but only a few leaders are willing to rip up the blueprint and build something new. For everyone else, the returns will stay just out of reach, hovering close enough to chase but never quite close enough to bank. What do you think, are big banks finally getting AI right, or still selling a dream? Drop your thoughts in the comments and share this story using #AIinBanking if it made you rethink where the industry is headed.
