Assumption Audit

When to use

Use before finalizing or presenting any strategy, business case, or major decision — especially one that has been worked on for a while by a small group, since that's when unstated beliefs calcify into "known facts." Also use as a dedicated pre-mortem step before launch, and any time someone says "we all agree this will work" without anyone having named why.

What it does

Extracts the assumptions a plan silently rests on (not just the ones people are willing to say out loud), classifies each by evidence quality, and identifies which ones are load-bearing — meaning if they're wrong, the plan doesn't just underperform, it fails. It then designs the cheapest test that would falsify each load-bearing assumption before real capital is committed.

Method

1. Reverse-engineer assumptions from the plan, don't just ask people what they assume. For every claim, number, or causal link in the strategy ("customers will pay $X," "channel Y will scale," "competitor Z won't respond"), ask "what would have to be true for this to hold?" — that's the assumption.
2. Classify each assumption's evidence base: proven (direct evidence from this business), analogous (evidence from a comparable situation elsewhere), or belief (no real evidence, just conviction or precedent-by-repetition).
3. Score each assumption on two axes: how much the plan's outcome depends on it (impact if wrong), and how confident the evidence base is (confidence). Plot on a 2x2 — high impact / low confidence is the danger zone.
4. Distinguish load-bearing from decorative. A load-bearing assumption, if false, breaks the plan's core logic (e.g., "the market will pay a premium for X"). A decorative one just changes a detail (e.g., "the launch color will be blue"). Focus scrutiny only on load-bearing ones — auditing every assumption equally dilutes attention from the ones that matter.
5. Run a pre-mortem on each high-impact/low-confidence assumption: "assume this failed — what evidence would we see, and could we have seen it earlier?" This surfaces failure modes faster than asking "what could go wrong" in the abstract.
6. Design the cheapest falsification test for each danger-zone assumption — a pilot, a customer interview round, a pricing test, a competitor-response scenario — sequenced by which one, if it fails, kills the plan earliest and cheapest.
7. Write the "if we're wrong" trigger: for each load-bearing assumption, name the specific signal that would tell the team it's wrong in time to change course, not after the capital is spent.