The decision, not the demo
Most AI pitches fail because they start with the tool. Stakeholders nod, ask for a pilot, and the thing dies in a shared drive.
Write the decision you want in one sentence before you open a slide. “Approve a 90-day content workflow trial with one owner and a kill date.” That is a decision. “Explore generative AI” is a wish.
If they cannot say yes or no in the room, you did not ask for a decision.
What good looks like in 90 days
Pick one production path. Content drafts, first-line support answers, or internal research briefs. Not three.
- Volume: how many units per week vs the current baseline
- Time: hours saved per unit, measured on real work, not a happy-path demo
- Quality: who signs off, and what “good enough to publish” means in writing
- Risk: what is allowed to go out without a human, and what never is
If you cannot name the baseline, you cannot claim a lift. Spend week one measuring the old way. It is boring. It is the only number a CFO will trust later.
ROI a finance person can argue with
Keep the math on one page. Cost of the old path. Cost of the new path. What you stop doing if it works. What you still pay if it fails.
| Line | How to fill it |
|---|---|
| Current cost | People-hours × loaded rate for the chosen path, last 90 days |
| New cost | Tooling + review time + one owner. Include the review. Always. |
| Upside | Hours back, or extra units shipped at the same headcount. Pick one. |
| Payback | Weeks until the new cost is lower than the old cost. If you cannot see it, shrink the scope. |
Do not forecast brand lift. Do not forecast “innovation.” Those are real. They are not this slide.
Name the risks first
Put the ugly list above the roadmap. People relax when they can see you already found the holes.
- Confidential data in prompts or vendor logs
- Confident nonsense published because nobody owned the review
- A model change that quietly wrecks the workflow in month two
- A team that treats the tool as extra work instead of a replacement for a step
For each risk: owner, trigger, and what you do. “We will monitor” is not a plan.
What you will not claim
Write the boundary in the pitch. It makes the rest of the claims easier to believe.
- We will not replace the people who currently own this work in 90 days
- We will not publish unreviewed customer-facing copy
- We will not feed unpublished strategy, legal, or HR files into a public model
- We will not scale to a second team until the first path has a measured baseline and a kill date
The 30-day next step
End with a sequence, not a vision.
- Days 1–7: measure the current path. One owner. One spreadsheet.
- Days 8–21: run the new path on live work, with the same reviewer as today.
- Days 22–30: compare units, time, and error rate. Keep, shrink, or kill.
Print this page. Fill the blanks in ink. If the blanks stay empty, you are not ready to pitch.