Most AI business cases fail for the same reason: they lead with the technology and leave the arithmetic for later. Finance teams read them the other way round. If you want a CFO to sign, write the case the way they will check it.

Start with today’s baseline

Before you estimate a saving, measure what the process costs now. Count volume per month, average handling time, error or rework rate, and the fully loaded hourly cost of the people involved. Two weeks of honest sampling beats any industry benchmark.

This step also protects you later. Once the system is live, the baseline is what you’ll be compared against — so agree it with finance up front, in writing.

Be specific about whose time you’re measuring. A saving spread thinly across forty people is real, but it rarely shows up in a budget. A saving concentrated in one team — enough to avoid a planned hire, say — is far easier for finance to recognise.

Estimate value in three lines

Keep the value model simple enough to fit on a napkin. Almost every credible case we’ve seen is built from three lines:

• Time returned — hours saved × loaded hourly cost × a realistic adoption rate (we use 60–70% in year one, not 100%).

• Revenue protected or gained — faster responses, fewer lost deals, fewer no-shows. Use your own conversion data, then discount it.

• Cost avoided — rework, penalties, overtime, or hires you no longer need to make.

Show the low, expected and high case side by side. A CFO trusts a range far more than a single confident number.

Be honest about cost and risk

Costs include more than the build: model usage, hosting, monitoring, internal time for testing, and training. Add a line for ongoing tuning — systems that aren’t maintained drift.

The question finance always asks: “What happens if it doesn’t work?” Answer it before they ask: define a checkpoint at week six with a go / no-go metric, and cap the spend until that point.

Then list the top three risks — data quality, adoption and accuracy — with a mitigation for each. Naming risks plainly builds more credibility than any projection.

Put it on one page

The final document should fit on a single page: the problem in one sentence, the baseline, the three-line value model, total cost, payback period and the week-six checkpoint. Everything else belongs in an appendix.

Finance doesn’t need to understand the model. They need to understand the money — and when they’ll see it.

For Vantage Capital, this format turned forty ideas into five funded projects worth an estimated $1.2M a year. The ideas weren’t better than anyone else’s. The arithmetic was simply easier to say yes to.