Pre-Mortem — what could go wrong before you commit?

First, in 30 seconds

A pre-mortem is a post-mortem you do before the decision.

Instead of asking "what could go wrong?" — which the optimistic brain waves away — you assume the plan has failed and ask "what killed it?" That single reframe surfaces the real risks. This tool then scores each one the way engineers and manufacturers do: Risk Priority Number = Occurrence × Severity × Detection (the FMEA method, MIL-STD-1629A / AIAG-VDA).

The output isn't a feeling. It's a ranked register, a hard verdict (proceed, proceed with guardrails, or fix first), and — most useful — a kill-criterion and leading indicator for each risk so you know in advance what would make you stop, and what early signal to watch for. Then you come back and track whether those signals fire.

Occurrence (O): How likely the failure is. Low = 2, Med = 5, High = 8.
Severity (S): How bad the impact is, 1–5, doubled to a 2–10 scale.
Detection (D): How hard it is to see coming. Visible = 2, partial = 5, blind spot = 9.
RPN: O × S × D (8–720). ≥200 act, 80–199 watch, <80 note.

The plan you're about to commit to

A decision, a launch, a hire, a raise, a bet. Be concrete about what success looks like.

The plan / decision

What are you committing money, time, or reputation to?

Success looks like

The outcome that means this worked.

Deadline

When the verdict is in.

Resources committed

Money, headcount, time — what's on the line.

Key assumptions

What must be true for this to work? One per line.

Pasted context (optional)

Notes, a brief, a doc — anything that helps the AI suggest risks (only sent if you use an AI button).

How it fails — the risk register

List the ways this plan dies. For each, set a kill-criterion (what would make you stop) and a leading indicator (the early signal). Add as many as you need.

Likelihood → probability map

Optional. The probability-weighted cross-check uses these. Defaults reflect typical base rates; adjust to your context.

Low =

Med =

High =

The verdict

The decisive risks

Stated as decisions. These are the ones that should change your mind, your plan, or your guardrails.

The failure register (FMEA)

Ranked by Risk Priority Number. O = occurrence, S = severity, D = detection, RPN = O×S×D. PW = probability-weighted impact, an independent cross-check.

Risk Priority Number by risk

A scored register is the start. A second read is the safeguard.

Founders are systematically blind to their own plan's blind spots — that's the whole reason the pre-mortem exists. If real money, time, or reputation is on the line, it's worth a calm outside read of your register before you commit. I do this with founders every week, and the first conversation is free.

Get a second read →

How the scoring works (for the curious)

This is FMEA — Failure Modes and Effects Analysis, the method used in aerospace and manufacturing (MIL-STD-1629A, AIAG-VDA). Each risk gets three numbers: O (occurrence, from likelihood: low 2, med 5, high 8), S (severity = your 1–5 impact × 2), and D (detection, from how visible it is: visible 2, partial 5, blind spot 9 — harder to see scores worse). The Risk Priority Number is RPN = O × S × D, ranging 8–720.

Bands: RPN ≥ 200 = ACT, 80–199 = WATCH, < 80 = NOTE. The probability-weighted cross-check is (likelihood% / 100) × impact — a second, independent lens so a single bad score doesn't drive everything.

The verdict tree: a risk is existential if impact ≥ 4 and likelihood is med or high. If any existential risk has no kill-criterion and no leading indicator, the verdict is FIX FIRST — you're flying blind on something that could end the plan. Otherwise, if any risk scores RPN ≥ 200, it's PROCEED WITH GUARDRAILS. Otherwise, PROCEED.

Every number here is computed in your browser from your inputs — none of it comes from AI. The optional AI buttons only help you brainstorm and phrase; they never touch the scores or the verdict.

A structured decision aid — not financial, legal, or investment advice. The scores reflect the inputs and judgments you enter; the method organizes your thinking, it doesn't replace it. Everything stays in your browser and saves locally on this device — nothing leaves except the optional AI calls you trigger. Logic current as of June 2026.