---
title: "Prompt: pre-mortem for an AI decision"
slug: pre-mortem-decision-ia
lang: en
updated: 2026-10-19
---

# Prompt: pre-mortem for an AI decision

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 proposal or memo the committee is about to approve. Use it before the team session, not instead of it: the value of a pre-mortem is in what the people who know the operation write down. Do not paste confidential data unless your company has approved it.

```text
Act as a chief operating officer with 15 years of experience who has watched technology rollouts fail for reasons nobody said out loud in the approval meeting.

Context I am giving you:
- Company and area: [e.g. distributor, 400 people, sales area]
- Decision about to be approved: [one sentence: what, for whom, how long it commits us]
- What makes it hard to reverse: [contract, integration, data, people]
- Who proposes it and who approves it: [roles]
- Proposal or memo, if any: [paste text or write "none"]

What I need: assume it is twelve months from now and the decision failed completely. Produce a pre-mortem with exactly this structure:
1. Eight reasons for the failure, each in one concrete sentence. Cover different angles: adoption, data, vendor, process, people, cost, customer, compliance.
2. For each reason, an early signal observable at 60–90 days (a data point, not a feeling).
3. For each reason, what we would change today, before approving.
4. The three reasons that, given the context, should worry us most, and why.
5. Three uncomfortable questions the group will probably avoid and that are worth asking in the session.

Output format: a table with the columns Reason · Early signal · What we would change today · Suggested owner (role, not name), followed by points 4 and 5 as bullets. Maximum 450 words. No introduction.

Rules:
- Do not invent figures or other companies' results. If a signal needs a threshold, write [THRESHOLD PENDING] and tell me how to set it.
- Mark every assumption as an assumption.
- No generic reasons ("poor communication"); each reason must name what failed and where in the process.
- 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:

- Is each early signal something you can measure or see on a date, or did it slide back to "the team is not engaged"?
- Is one of the eight reasons the one your team does not dare to say? That one goes first in the session.
- Do the "what we would change today" items fit before approval, or are they tasks for later? If later, they are not useful.