---
title: "Prompt: one-page AI business case"
slug: caso-de-negocio-una-pagina
lang: en
updated: 2026-10-19
---

# Prompt: one-page AI business case

How to use it: copy the block, paste it into ChatGPT, Claude or Gemini and replace the fields in brackets. It works best if you also paste the vendor's proposal and the pilot report, even as a draft. Do not paste confidential data unless your company has approved it.

```text
Act as a chief financial officer who has reviewed dozens of technology proposals and who rejects any benefit that does not distinguish what was measured from what is hoped for.

Context I am giving you:
- Proposed decision (what, for how long, who signs): [ ]
- Baseline measured before: volume, time per task, review, exceptions, date and sample: [or write "no baseline"]
- Pilot results, if any: [ ]
- Known costs: licences, integration, supervision, training, exit: [ ]
- Vendor proposal or savings figure: [paste text]

What I need: write the business case on one page with these seven blocks, in this order:
1. Decision in one sentence.
2. Baseline.
3. Costs that go in (all five, even if one is zero).
4. Verified benefit.
5. Estimated benefit, with the assumption written down.
6. Assumed benefit, kept separate.
7. Sensitivity: what happens if the benefit is half.
Calculate net hours released as: volume × (baseline minutes − new execution minutes − review minutes) / 60 − additional exception hours. Close by answering the five questions a CFO would ask: what baseline and sample support the benefit; who reviews and handles exceptions and what it costs; which cost actually changes or where capacity is redeployed; what happens with lower adoption or half the benefit; what evidence would make us stop the plan and what it costs to exit. Where you cannot answer with the data I gave, write "evidence gap".

Output format: a table Block · Content, followed by the five questions with brief answers. Maximum 450 words. No introduction.

Rules:
- Do not turn released hours into money saved unless I tell you which real cost falls.
- Do not invent figures. If data is missing, write [DATA PENDING] and tell me how to get it.
- Label every assumption as an assumption.
- If critical information is missing, ask me at most 3 questions before answering.
- Answer in plain, direct business English, with no sales adjectives.
```

What to check in the answer:

- Did it put anything under "verified" that was not actually measured with the same scope as the baseline?
- Does the sensitivity change the recommendation, or does it just repeat the number divided by two?
- Do the five questions have answers backed by data, or "evidence gap" where appropriate? If all five have answers, be suspicious.