market-sizing git:20260728.2d6d1c9A

Immutable. This exact content is served forever at /api/v1/blob/2b3f7bb05ff8f9a2.

---
name: market-sizing
description: Estimate how large an opportunity actually is, from the bottom up, and state the assumptions that drive the number. Use when evaluating a market, writing a plan, or challenging a large number in a deck.
---

# Market sizing

Most market sizes are top-down and useless: a large industry figure
multiplied by an invented percentage. A bottom-up estimate is smaller,
defensible, and actually informs a decision about whether to build
something.

## Method

1. **Build from units and price, not from a total.** How many potential
   buyers, how many will realistically buy, what they will pay, and how
   often. Each step is arguable, which is the point.
2. **Distinguish the three sizes.** Everyone who could conceivably buy,
   the segment you can actually serve, and what you could realistically
   capture in a few years. Conflating them is the standard deck error.
3. **Cross-check top-down against bottom-up.** Wildly different answers
   mean an assumption is wrong, and finding which is more valuable than
   the number.
4. **State every assumption with its source.** A number whose
   assumptions are visible can be argued with and improved; a single
   figure cannot.
5. **Sanity-check against existing players.** If your estimate implies
   the incumbents should be several times larger than they are, the
   estimate is wrong.
6. **Size the beachhead, not the dream.** The immediately addressable
   segment determines the next two years; the total market determines
   nothing operational.
7. **Test sensitivity.** Which assumption most changes the answer is
   where to spend research effort (see risk-analysis).

## Boundaries

Sizing estimates opportunity, not winnability, which depends on
competition and execution (see competitive-strategy). Precision is false
comfort; the useful output is an order of magnitude with named
assumptions. New categories cannot be sized from existing data and need
analogues instead.