One Big Idea

Two banks buy the same AI — one earns five percent, the other fifty. The difference was never the model. It was the operating model around it.

Same models, same vendors, roughly the same budget — and one bank earns five percent while the other earns fifty. The one thing money can buy, the model, is the thing becoming free: the same frontier models sit behind the same cloud interfaces, available to every competitor. So the model is no longer the edge. What separates the bank that earns five from the bank that earns fifty is the machinery around it — how the work is redesigned, how freed capacity is converted, and how the whole thing is governed. In the Boardroom Equation, this was the multiplier term. This issue opens it up.

The Insight

Think of realized AI value as a simple product: raw capability × M, the operating-model multiplier. Raw capability is what you buy — nearly identical across competing banks. M is what you build: below one at the bolt-on bank, where the operating model leaks most of what the AI produces; two or three times at the rewired bank. And M is a product, not a sum — set by three levers that multiply, so zero on any one lever is zero overall.

This is why a reported seventy-three percent of banking AI pilots stall before production: not because the models failed, but because the operating model around them never changed. The capability went up; the multiplier stayed at one.

What you buy is not what you bank. The model is nearly identical across competitors; the multiplier is the only thing you build.

Framework of the Week · The Operating-Model Multiplier

M is set by three levers that multiply — so the rule writes itself: when the multiplier sits at one, hunt the broken lever, don't buy a better model.

  • 1 · Rewire — redesign the process around the AI instead of bolting the AI onto the process. Bolt-on returns a reported 5–15%; rewiring returns 20–50% that compounds. The signature is BCG's 70/20/10: ~70% of effort on people & process, 20% on the data & technology backbone, 10% on the model itself. Most banks deploy the model and invert it.
  • 2 · Reallocate — freed time is not profit until you reclaim it. By some estimates, up to two-thirds of AI-freed time leaks back into low-value work unless capacity is redeployed or removed. When capacity has nowhere to go, leaders should not book it as return.
  • 3 · Re-govern — after SR 26-2, governance is an enterprise operating-model layer, not a model-risk silo — the rail that lets revenue and agentic AI scale safely, not the brake.

The fastest way to read a bank's multiplier is to read where the money goes. A high-multiplier bank spends most of its AI effort on workflow redesign, capability, and governance — the model is a small line, and its pilots reach production because the process was rebuilt to receive them. A low-multiplier bank spends most on the model and the software, treating people and process as a footnote — its pilots stall, and a measured efficiency gain barely moves return on equity.

The full framework, with the multiplier equation and the 70/20/10 signature drawn out, lives in the Frameworks library.

Use Case · Reading the 70/20/10 signature

The same logic grades by use case. A copilot pays only if headcount actually moves; fraud detection and credit underwriting pay when the workflow and governance around the model are rebuilt. Read the ratio, and you can predict the return before a single model is chosen: the bank spending 70% on people and process is heading for the compounding 20–50%; the bank spending 70% on the model is heading for the stalled pilot and the flat return on equity.

Risk Note

Two cautions. Agentic AI raises the stakes — BCG projects autonomous agents rising from ~17% of AI value today to ~29% by 2028, across industries, and agents are far less forgiving of a weak operating model than a chatbot. They need clean data, redesigned workflows, and enterprise governance to function at all. And the model is not the moat — as frontier models commoditize, out-purchasing competitors buys nothing durable. The only edge that compounds is the operating model — the one thing that cannot be bought off the shelf.

Latest Video

This week's video — Why Two Banks Buy the Same AI and Only One Gets the Return — walks the multiplier equation, the 3 levers (Rewire · Reallocate · Re-govern), the 70/20/10 Law, and the Multiplier Audit you can run on Monday: measure M → find the broken lever → re-weight to 70/20/10 → lock the capacity.

Watch: youtu.be/uHQYr-BhYGU

The free five-page playbook in the Frameworks library turns it into a worksheet for your own AI portfolio.

Reply and tell me which of your three levers is sitting at zero — that's the one capping your AI return. I read every response. Forward this to a banking executive whose pilots keep stalling before production.

Was this forwarded to you? Subscribe to The AI Architect Letter — free, every Saturday.

Minh Tran · AI Business Architect · LinkedIn · Workshops & advisory: aibusinessarchitect.ai