Numbers without a data team
Lesson 17 of 24, level 01Index
“How big is this?” Nobody will run a query for you this quarter.
Estimate out loud, so people argue with the assumption
Make the estimate inspectable
- Population: Eligible users
- Frequency: Jobs per month
- Value: Value per job
- Range: Low / mid / high
A Fermi estimate breaks a number you cannot know into numbers you can guess, then multiplies. Its value is not precision, it is that every assumption is visible and therefore attackable.
Which is the point. A sized opportunity with three stated assumptions produces a useful argument. A confident number with no workings produces either blind agreement or blind refusal.
A rough estimate is useful when it exposes the inputs that matter. It becomes dangerous when a precise-looking total hides guesses, overlapping populations, or incompatible units.
Use your chosen course project throughout. The additional examples below are fictional practice cases; transfer the method to your own evidence.
Build a small equation
Break the opportunity into eligible people, frequency, and effect per event. State the period and the definition of eligibility. Avoid adding overlapping groups without deduplication. A bottom-up model tied to your product is often easier to inspect than an impressive total addressable market number unrelated to this decision.
Use ranges honestly
Give uncertain inputs a low, base, and high assumption with a reason. Do not call these confidence intervals unless you actually calculated them using a suitable method. Change one input to see what dominates the result. That tells you which missing fact is worth researching before spending more time on the model.
Connect size to a choice
An opportunity estimate does not prove your solution will capture the opportunity. Separate the affected population from adoption and from the expected improvement. Ask whether a conservative case would still justify a small investigation. If only the most optimistic case works, the next action should resolve the fragile assumption.
Size the reachable change
- Eligible: Who faces the task?
- Frequency: How often?
- Adoption: Who uses the fix?
- Effect: Change per event
Doing it with AI, and where it breaks
Have a model build the Fermi estimate structure, then check its arithmetic and, more importantly, its priors.
Its priors come from the average of the internet, not from your market. It will assume US pricing, US conversion rates and US salaries unless you tell it otherwise.
Your AI workbench
Start with your own notes or clearly labelled practice data. Remove private details before sharing. Replace the placeholders, run the prompt in your chosen AI tool, and keep the output beside its source.
Build a bottom-up sizing model for [project] using only these supplied inputs [paste]. Show units, formula, low/base/high assumptions, overlap risks, and adoption separately. Do not invent market data. Identify the single input whose verification would most improve the decision.
Before you use the output
- Units cancel correctly
- Adoption is separate from eligibility
- Ranges are assumptions rather than claimed certainty
Stuck? Try this next
Ask AI to calculate a one-user, one-event example first. Scaling a wrong unit conversion across a large population only makes the wrong answer more persuasive.
Keep a brief AI log: input used, useful output, what you checked, and what you rejected. The decision remains yours.
Build it
A sized opportunity with every assumption written down and attackable.
Checkpoint
If you did the build, these take two minutes. If you cannot answer one of them, that is the part to go back to.
Moving on marks this lesson complete. Finish the build first, it is the part that counts.