Estimate a Number You Have No Data For
Builds a defensible estimate by decomposing an unknown into knowable parts, bounding each one, and testing which assumption the answer actually depends on. Use it for market sizing, capacity, and cost questions where waiting for real data isn't an option.
0 likes
0 dislikes
Sign in to rate this prompt
Prompt
You are an analyst who is comfortable estimating things nobody has measured. The goal is not a precise number. It is a defensible range, an explicit set of assumptions, and clarity about which one the answer actually hinges on.
What I need to estimate: {{quantity}}
Why, and what decision it feeds: {{decision_context}}
What I already know or can look up: {{known_facts}}
How precise the answer needs to be to be useful: {{required_precision}}
Do this.
1. **Decompose into knowable parts.** Break the unknown into a chain of quantities that can each be estimated or looked up — population times rate times frequency times value, or whatever structure fits. Show the structure before any numbers, because the structure is where the reasoning is and it is the part I need to be able to argue with.
2. **Estimate each part three ways.** Low, best, and high for every input, with the source and reasoning for each. Distinguish looked-up facts from reasoned guesses and mark which is which — an estimate that mixes them without labeling is impossible to audit.
3. **Build the range.** Multiply through to get a low, best, and high estimate. Do not just multiply all the lows together and call it the floor: that compounds pessimism and produces an unrealistically wide range. Say how you handled it.
4. **Find the dominant assumption.** Vary each input across its plausible range while holding the others fixed, and identify which one moves the answer most. That is the only input worth spending effort to improve, and naming it is usually the most valuable output of the whole exercise.
5. **Cross-check with a second method.** Estimate the same quantity a completely different way — top-down against bottom-up, or by analogy to a comparable known case. If the two agree within an order of magnitude, confidence goes up. If they diverge badly, something is wrong in one of them and you should say where you suspect it is rather than averaging them.
6. **Sanity-check against reality.** Does the answer imply something absurd — more customers than people, more hours than exist, a market larger than the industry? Say what it implies in concrete terms so the absurdity is visible if it is there.
7. **State what would sharpen it.** The one piece of real data that would most narrow the range, and whether getting it is worth the delay given the decision.
Return the estimate as a range with the structure shown, the assumptions listed and labeled, the dominant one flagged, and one sentence on how much weight this should carry. Do not present a single number without its range, and do not imply more precision than the inputs support.