Measuring and funding AI with the finance function.
A series for CFOs and Chief AI Officers on baselines, attribution, inference cost and stage-gated funding.
Boards keep asking what their AI spending has returned, and most management teams cannot answer with a figure they would defend in front of an auditor. The cause is rarely the technology. Nobody recorded the starting point, nobody agreed how to separate the AI's effect from everything else that changed, and the running cost of the models sits in a cloud bill nobody maps to a use case.
Those are finance questions as much as AI questions. This series sets out a method the CFO and the Chief AI Officer can run together, from the baseline taken before a pilot starts to the gate that stops funding when the pilot misses its target.
Written by a finance practitioner and an AI governance practitioner.
Each part is co-written with Dr. Christina Chua, so the method has to make sense to both the person who funds the work and the person who runs it.
Numbers come with their source and grade.
Every statistic is graded as measured, reported, forecast or our own judgement, so you can see what each claim rests on.
Every part ends with something to do.
A template, a gate criterion or a question for the next investment committee.
The parts
1 of 6 published
A Joint CFO and CAIO Framework for AI Benefit Realisation
Most AI programmes cannot say what they returned because nobody wrote down what they were meant to return, or what they cost to run. The finance function and the AI function each hold half of the answer.
What you take away Four controls the CFO and the Chief AI Officer sign together: a baseline, an attribution method, a unit-cost line for inference, and a stage-gate that stops funding.
Read article →Setting the Baseline Before an AI Pilot Starts
A baseline taken after the pilot starts is a guess. What to measure, for how long, and who signs it off before the first model call.
What you take away A one-page baseline record that finance will accept, with the variance a pilot must beat to count.
Planned
Attribution Methods for AI Benefits
Revenue rose and the AI went live in the same quarter. Holdouts, staggered roll-outs and difference-in-differences separate the two, and each has a cost.
What you take away Which method fits which use case, and what to do when no control group is possible.
Planned
Inference Cost Accounting for Finance Teams
Per-token prices fall every quarter while the bill rises. Unit economics for AI, tagged to the use case that ran it.
What you take away A cost-per-outcome line for each AI use case, and the three numbers that make a monthly AI cost review useful.
Planned
Stage-Gates for AI Pilot Funding
Pilots are cheap to start and hard to stop. Gates with kill criteria written in advance, and the governance that makes stopping normal.
What you take away Gate criteria, decision rights and a template for the stop decision.
Planned
Reporting AI Benefits to the Board
What a board pack should show about AI value: realised against forecast, net of run cost, with the confidence of each figure stated.
What you take away A one-page AI benefits report that survives an audit committee.
Planned
Christina and I came at this from opposite ends. She has spent years watching new products reach clients through finance, compliance and technology teams that measure success differently. I have spent years on infrastructure programmes where nobody gets the next tranche of money without showing the last one worked. AI spending has mostly skipped that discipline. This series is our attempt to put it back without slowing the work that deserves to go ahead.
New parts by email
Each part goes to the newsletter when it is published. If you would rather talk about a specific AI investment decision, the other door is open.