The AI business case template

Here is the template: seven fields, each with a one-line prompt for what goes in it, pre-loaded against the five reasons AI proposals get rejected. No download, no email wall – the whole thing is on this page. Copy it into a document and fill it top to bottom; the only part that takes real work is the financial section, and there is a worked example below.

What's in the template

Seven fields. Each line tells you what goes in it:

  1. The problem, in dollars. What the status quo is costing per year – a counted number, not an adjective.
  2. The proposed pilot. One process, one team, one quarter: what changes on day one, and what is explicitly out of scope.
  3. Costs. Software, staff hours at loaded rates, and any outside help. Nothing discovered later.
  4. The benefit, as a range. Hours or errors avoided, priced out, assumptions listed. See the worked example below.
  5. Owner and timeline. A name (not a team), a reporting rhythm, and three dates: start, checkpoint, decision.
  6. Risks and the exit. Three real risks with mitigations, then the checkpoint, the metric that rules it, and what shuts down on a miss.
  7. The ask. Amount, duration, decision requested – one sentence an approver can say yes to.

What separates this from a generic business-case template: every field is aimed at one of the five ways AI proposals actually die – the missing number, the missing owner, the funded competitor, the threatened territory, the missing exit. Fill all seven and there is no empty space for a no to land in.

How to fill the financial section

The financial section is fields 3 and 4, and it is the only part that takes longer than an hour. Here it is filled in for a real-shaped example – an accounts team piloting AI-assisted invoice matching:

  • Costs: $3,600 software (12 months), $5,000 in staff time (100 hours loaded at $50), no outside help. Total: $8,600.
  • Benefit: 30 matching errors a month, an hour of rework each, is 360 hours a year. Priced at the same loaded $50, that is $18,000 a year at full catch rate – stated as $11,000 to $18,000, assuming the pilot catches 65 to 100 percent of errors. Assumptions named: error count from the Q2 exception log, rework time from the team lead, loaded rate from finance.

Two things to copy from the example even if your numbers are different: the range is honest about adoption, and every input has a named source. A reviewer who can see where each number came from argues with your sources, not your credibility. And check the floor: if the bottom of your range does not clear your costs inside a year, pick a different pilot.

Your First Move

See where your organization stands on AI – in 3 minutes.

Five questions. One clear picture across all five dimensions of AI readiness. No credit card, no sales call, no fluff.

Take the QuickScan →

Free · 5 questions · 3 minutes · Instant results