Lesson 06 of 20
Build the AI-Adoption ROI Case That Survives the CFO
Internal Reporting & Value-Proof · Cross-category
Synthetic case data — evidence from real ChatGPT, Claude, Gemini and Copilot runs.
✓ Every AI output quoted below is from a real run on 2026-07-02 — screenshots in evidence/. ChatGPT, Copilot, and Claude each ran the full sequence (naive prompt → defensible brief → compression to one page); the showcases below use all three on the naive run and Copilot + Claude on the brief and compression. Gemini is not part of this lesson's validated evidence — treat any note about it as a check to run, not a result. Every figure in the lesson's own target one-page brief is independently checked by verify_arithmetic.py in this lesson's folder; the numbers attributed to each tool are quoted from its transcript and screenshot.
Your VP, Tomás, forwards a one-line note from the CFO — Diane Okafor: "Need the AI pilot numbers. One page. Thursday." No template. No format. Just prove that the thing you have been championing paid for itself.
You believe it did. You watched six people cut their task times roughly in half. But "it felt faster" is not a number a CFO signs, and your instinct right now is pulling you toward the biggest, roundest figure the data will give you.
That instinct is the whole lesson. The biggest number the data gives you is not the number that survives Thursday.
Three files are on your desk. Open all three before you start — the answer lives between them, not in any one.
- File 1 — the task-time log. Six pilot members tracked before-and-after times on every AI-assisted task over twelve weeks. Six task types, from quote comparisons to the annual strategy write-up.
- File 2 — the review log. Started in week four, after an AI-generated output went to a stakeholder with a fabricated supplier-news item in it. It tracks how long people spent checking and correcting AI output before it was safe to send.
- File 3 — the cost summary. Licenses, training, the coordinator's hours. And one number most people miss: the loaded hourly rate — $68/hr ($48 salary + 42% benefits burden), not the salary figure.
Your deliverable is one page Diane can approve in ten minutes. Not a dashboard. Not a deck. One page.
You’ve read the free three. The other seventeen are where the misses live.
The first lessons show what a general-purpose AI catches on a procurement task. The rest show what it misses — the unit conversion that survives an expert prompt, the two clauses it lists but never connects, the table that contradicts its own arithmetic. That gap is the job.
Full course access
For individual practitioners
- ✓All 20 lessons, permanent access
- ✓Practice case files for every lesson
- ✓Real ChatGPT / Claude / Gemini / Copilot evidence
- ✓New lessons included as they ship
Live team workshop
For procurement teams and their leaders
- ✓Live 2-hour workshop, run for your team
- ✓Built on the same 20-lesson curriculum
- ✓Your categories, your workflows
- ✓Led by a practitioner running AI adoption inside a Fortune 500 procurement organization