AI / Analysis & Decision Making
Unpack My Assumptions
Find the premises the conclusion never states
Here is my reasoning: [reasoning]. Please: 1) List every implicit assumption in it (premises that are never stated but must hold); 2) Mark the most fragile one; 3) If that assumption does not hold, how does the conclusion change; 4) What information would make me more certain.
How to use
Ask it to quote the sentence each assumption comes from; assumptions it cannot locate are usually its own invention.
Good for
- The reasoning reads smoothly and every step “obviously holds” — smooth reasoning usually means the most fragile assumption is hiding in a transition sentence.
- Before you use this reasoning to persuade someone or ask for budget, self-check: which single premise, if false, collapses the whole thing.
Not for
- The facts the reasoning rests on aren't verified yet (unclear source, unclear definition) — unpacking assumptions looks rigorous, but the whole structure sits on a wrong fact.
- You just need a rough number or a throwaway decision — unpacking assumptions costs more than the decision is worth.
Example
Input
We think support can go from 12 people to 6: since we launched the AI agent, human ticket volume is down 55%, and the remaining staff can handle it.
What you get
You want the two unstated premises named — “volume stays at the new level after the 55% drop” and “the remaining six handle the same mix of tickets” — with the weaker one flagged (complex complaints were only deferred; cancellations and bad reviews come back in about two months). It should end by saying what data would settle it: handling time broken out by ticket type, not total volume.
Common mistakes
- Turning “list the implicit assumptions” into “list everything that could go wrong” — you get fifteen generic risks and the fragile one drowns.
- Stopping at the list and never revising the reasoning: a beautiful inventory, and the conclusion is untouched.
Try tweaking it like this
- Add “rank them by whether the conclusion reverses if this one fails” — the list turns from caveats into the actual fork points.
- Change “my reasoning” to “my reasoning, plus the one question I least want to be asked” — that question usually points straight at the weakest assumption.