Somebody is going to hand you an AI number this quarter, and it will be gross. Hours saved, tickets deflected, two days off the close, measured in the unit that flattered it by the team that wants the next tranche. Your job is the net, and no other function in the company can compute it. This course teaches the CFO's side of AI as a finance discipline rather than a technology brief: the total cost of completing a process before and after, with the errors and delays the change created sitting inside the after figure; where agent cost actually migrates to and whose budget absorbed it; which costs you are allowing to become permanent and what operating leverage does to you in both directions; how to win the budget by changing which number the room is arguing about; the build-versus-buy fork and the vendor questions that decide it; where agents genuinely pay, which is working capital rather than headcount; and what a buyer or a board can actually inspect. Ten short lessons, finance chief to finance chief, and you finish holding two things: a net-benefit sheet that is allowed to come out negative, and a CFO AI scorecard with an owner and a cadence on every line.
Somebody is going to hand you an AI number this quarter, and it will be gross. Hours saved. Tickets deflected. Two days off the close. It will arrive in the unit that flattered it, from the team that wants the next tranche, and it will be broadly true. Gross is not a lie. It is a partial. The net is the part nobody was assigned, and computing it is now the finance chief's job, because no other function sees the exception queue, the platform invoice, the integration hours and the headcount that failed to leave in the same place.
This course treats AI as a capital allocation rather than a technology programme. You will learn the method that produces a defensible per-unit number, including the term that lets the arithmetic come out negative. You will trace where agent cost actually lands, which is frequently a budget that is not yours. You will decide which AI costs you are allowing to become fixed, and what that decision does to you when volumes move in the wrong direction. You will prepare a budget argument that wins on operational arithmetic instead of a promised return date, and you will build the evidence a buyer or an audit committee can inspect without asking you for it.
The deliverables are two. The AI net-benefit sheet: one row per initiative, the benefit as reported, the total cost of completing that work afterwards, and which cost centre absorbed each line. And the CFO AI scorecard: three layers, one owner and one cadence per line, plus an explicit list of the numbers you have decided not to report.
CFOs and finance chiefs who now sponsor and answer for AI investment rather than approving it from the sidelines.
VPs of Finance, controllers, and FP&A leaders who are being asked to build the case and will be asked later whether it was real.
Founders and COOs who own the numbers and need a return that survives a board, an auditor, and eventually a buyer.
A method for AI return that puts errors, rework and delay inside the after-figure, so the arithmetic can come out negative.
A cost trace that finds the integration, review and change-management spend sitting in budgets outside the AI programme.
A clear split between the AI costs that are genuinely fixed and the ones that scale strictly with volume, and what a contract does to that split.
A budget argument built on what the money buys down and what it retires, with the trigger for each tranche written before the money moves.
The build-versus-buy fork decided on differentiation and data defensibility, with the vendor questions, the red flags, and the lock-in test.
The working-capital case for agents, which is usually larger than the headcount case and almost always unmeasured.
A quarterly AI addendum a board or a buyer can inspect, and the discipline that stops four benefits being counted as one number four times.
The completed CFO AI scorecard, with an owner, a cadence, and a do-not-report list.