The capacity map shows where agentic AI creates capacity. This review — the loop's banking instrument — asks the CFO-organization question: where does it bank — on which line, signed by whom, visible in which ratio?
Why this is a business-transformation problem rather than a technology one — the three facts that set the frame, the four currencies value can be paid in, and the five levers it lands on — is argued once, in the standing case. None of it changes when the quarter does.
Set annual revenue and expense growth; the blue path recomputes off the Q2 2026 run-rate. The gold line is what the selected delivery scenario produces on the same axes — growth arithmetic is what the bar requires, the arc is what the program delivers. The arc has no 2026 point on purpose: the plan opens in Q1 2027, and nothing is assumed to ship this year.
Take capacity created by AI — hours per week — and split it: released to the expense line, or redeployed into growth work. Both are value; only one is "savings." The split is the management decision this whole map turns on.
Which line the released side lands on — and who owns it — is set by the five levers.
Scorecard metric this feeds: hours redeployed vs. released — the first-class measure that keeps capacity claims honest. For scale: Citi's disclosed ~100K weekly developer-hours ≈ 2,500 FTE-equivalents of capacity — value that banks only when a decision routes it somewhere.
One scorecard read at three altitudes — serving the earnings narrative, the operating review, and the investment committee. The weekly 100-process review is the natural home of the middle altitude; quarterly, the scorecard reconciles capacity claimed with P&L booked.