An AI proposal usually arrives at the CFO’s desk full of enthusiasm and short on the specific numbers a finance leader is trained to look for.
That is not a personality clash between technical optimism and financial caution. It is a mismatch in what each side considers evidence.
Short answer: a CFO evaluating an AI proposal is asking about cost, risk, and reversibility, not about the technology. Answer those three categories with numbers and the proposal moves. Answer them with adjectives and it sits in review indefinitely.
Why the standard pitch stalls
Most AI proposals lead with capability. What the system can do, how impressive the demo was, what competitors are already doing. A CFO hears all of that as unverified upside and starts looking for the numbers that were left out.
The proposal is not being rejected. It is being deferred until the missing numbers show up, which often reads as a no even when it is not one.
The questions that actually get asked
What does this cost, fully loaded
Not the subscription fee alone. Build time, review time, and ongoing ownership hours, converted to a dollar figure. A proposal missing this gets sent back before anything else is discussed.
What is the baseline we are improving on
A number captured before the project started, not an industry average. If this does not exist yet, the honest answer is that it needs to be captured first, which is a smaller and easier ask than the full proposal.
What is the payback period
Calculated from the fully loaded cost against the captured baseline. Under twelve months for a first project is a reasonable target to defend in the room.
What happens if it does not work
A stopping condition and a number below which the project gets adjusted or cut. This question is really asking whether you have thought past the launch date, and a clear answer builds more trust than any capability demo.
Who is accountable after it launches
A name with allocated hours, not a department. This is the single question most proposals fail to answer, because most proposals are written by the person excited to build it, not the person who will operate it.
What is the exposure if it makes a mistake
What data it touches, what decision it influences, and what the review process catches before it reaches a customer or a regulator. A specific answer here is worth more than a general assurance of safety.
Answering all six in one page
| Question | What satisfies it |
|---|---|
| Fully loaded cost | License, build hours, review hours, ownership hours, all priced |
| Baseline | Two to four weeks of real measurement, not an average |
| Payback period | A specific number of months, calculated from the two above |
| Stopping condition | A pre-agreed number and a decision date |
| Owner | A named person with allocated hours |
| Exposure | What data and decisions are involved, and what catches errors |
A proposal that answers all six on a single page moves through finance review faster than most capital requests, because it has already done the work a CFO would otherwise have to do themselves.
What CFOs are actually protecting against
Not AI specifically. The same pattern that sinks any project: an enthusiastic sponsor, no operational owner, and a return that was estimated rather than measured. AI proposals get extra scrutiny because the category has produced more of that pattern recently than most others.
FAQ
Should the CFO be involved before the project is scoped?
Ideally yes, at the readiness and use case selection stage, so the proposal that eventually arrives already reflects their concerns.
What if we cannot answer the exposure question well?
That is itself an answer. It usually means governance has not been discussed yet, which should happen before the proposal goes to finance, not during the review.
Does a strong answer to all six guarantee approval?
No, but it removes the objections that are actually about process rather than about the specific project, which is most of what stalls proposals in practice.
How different is this from a normal capital request?
Not very. AI proposals are held to the same standard as any other spending decision, which is often the gap between how they are pitched and how they are received.
Who should write the fully loaded cost estimate?
Whoever will operate the system day to day, with finance reviewing the assumptions. Estimates written by the vendor alone tend to underweight internal hours.
Where to go next: Before the next AI proposal goes to your CFO, answer the six questions above on one page. It is usually the difference between a quick yes and a proposal that sits for a quarter.
