AI Product Studio: The Org Design Question Most Banks Are Not Asking Out Loud

Most enterprise AI portfolios in financial services are not converting capital and headcount into shipped capability. The numbers are public. BCG's 2025 AI Value Study found that the gap between AI leaders and laggards is widening rather than closing, with the leaders capturing a disproportionate share of the value, and that the determining variable is not investment level. It is whether the institution has redesigned the way work gets done around the technology.

The reasons this is hard inside a regulated financial institution are structural. This piece is about what those structural reasons are, why they keep producing the same pattern of pilots-without-production, and what a different operating model looks like.

Where enterprise AI actually fails

Not at the model layer. The frontier model providers (Anthropic, OpenAI, Google DeepMind, and the open-weight community led by Meta and Mistral) have produced systems that are dramatically more capable than they were eighteen months ago. The models are not the bottleneck for any serious BFSI program in 2026.

It does not fail at the capability layer either. The largest North American banks (JPMorgan Chase, Bank of America, Wells Fargo, RBC, TD, Scotiabank) each employ hundreds of data scientists and machine learning engineers, with dedicated centers of excellence that did not exist five years ago. The same is true at the leading insurers: Sun Life, Manulife, Allianz, AXA, Allstate. These institutions have AI capability. McKinsey's 2025 Global Banking Annual Review counted more than 160 active agentic AI use cases across 50 of the world's largest banks. The capability is being built. The constraint has moved.

The failure is at the seam. Specifically, at the seam between the people who decide what an AI system should do and the people who know how to build it.