Over the past year, dozens of AI project justifications have crossed my desk — ours and our clients’. And I have noticed a pattern: the prettier the deck, the worse the project’s fate. Slides about “revolution” and “industry transformation” collect applause at the demo and die on the CFO’s floor. Because a CFO does not buy revolutions. A CFO buys a governed, measurable result on one process — with the ability to repeat it.
A good AI business case is not a presentation. It is a document after which the CFO calmly signs. Over years of working with enterprise clients we have settled on a six-block structure, and I am sharing it as is.
Block one — the process. Not “we are implementing AI” but “we are accelerating contract approval” or “we are reducing manual processing of incoming mail.” If the first line of the document does not name a specific process, you can stop reading — it is an application for an experiment, not for a result.
Block two — the baseline. How long the process takes today, where the errors occur, how many manual touches there are, who owns it. This is the most labor-intensive block and the one most often skipped. A mistake: without a baseline, any future “effect” is unprovable by construction — there is nothing to compare against.
Block three — the agent’s role. What it reads, what it proposes, what it executes itself, what it escalates to a human — and at which point a human confirms the decision. To a CFO, the phrase “the agent will handle everything” means “the boundaries of responsibility are undefined,” and the CFO is right.
Block four — the security loop. Which data stays inside the perimeter, which actions are legally significant, who has access. For regulated business this block is not fine print in the appendix but a condition of the project’s existence — I covered it separately.
Block five — the metrics. Cycle time, errors, SLA, risk, governability. How to measure the effect — and why “hours saved” is a weak metric if the process itself has not changed — I wrote about in detail here.
And block six — the plan: diagnostics, pilot, measurement, scaling. In exactly that order. A pilot without a pre-agreed measurement method is a postponed argument about what counts as success. Industry statistics are merciless: 41 percent of failed AI projects die because success criteria were invented after the fact.
Now, what must never go into a business case. First — “we will save 40%” with no source for the number: one unfounded figure devalues all the others. We enforce a hard rule internally: if a number is not confirmed, the document honestly says “from a case study” instead of a round number out of thin air. Second — “we will replace employees”: it is both inaccurate and toxic for an implementation that needs those very employees as allies. Third — comparisons with competitors without an evidence base: the CFO checks one random claim, finds a stretch — and stops trusting the entire document.
One last observation, from sales practice. A properly built business case travels well — it already contains the pain, the process, the owner, the metrics and the next step. It turns marketing interest into a qualified conversation. That is why we published our ROI worksheet in the open — fill it in for one of your processes and you will be holding a draft of blocks three, five and six.
The CFO is not the enemy of AI projects. The CFO is their best filter: he screens out what should never have been launched. Give him a document with a process, a baseline and a way to measure the result — and the signature will appear sooner than you expect.
