Market-Sizing Questions

Market-sizing questions — "how many gas stations are in the US," "what's the total revenue of the coffee shop industry in your city" — are Fermi estimation applied specifically to markets: sizes, revenues, unit counts. The interviewer already knows there's no way for you to know the real number; what they're grading is whether you can break an unfamiliar quantity into a chain of smaller quantities you can reason about, state your assumptions out loud, and land on a number in the right ballpark.

The structure: decompose into a chain of multiplications

Almost every market-sizing question reduces to the same shape:

$$\text{Target} = \text{Population} \times \text{Rate}_1 \times \text{Rate}_2 \times \cdots$$

In plain English: start from a population you have some grip on (a city's residents, a country's households, total car owners), then multiply by a chain of "what fraction of these actually do X" rates, each one individually easier to guess than the target number itself. The skill isn't guessing the final number directly — it's finding a decomposition where every single factor is something you can defend with a round, sensible estimate.

Worked example 1: coffee shop revenue in a city of 2 million

Break it down: assume 2 million residents, of whom roughly 40% (800,000) are regular coffee drinkers who buy from a shop rather than making coffee at home. Assume each buys coffee 3 times a week on average, at $4 per cup. Weekly revenue:

$$800{,}000 \times 3 \times 4 = 9{,}600{,}000$$

i.e. $9,600,000 per week. Annualize: $9{,}600{,}000 \times 52 \approx 500{,}000{,}000$, i.e. roughly $500,000,000 per year. Every assumption in that chain (40% coffee drinkers, 3 cups a week, $4 a cup) is individually defensible even though nobody has the real figure memorized — that's the entire point of the exercise.

Worked example 2: piano tuners in Chicago

Chicago has roughly 2.7 million people, or about 1 million households. Assume 1 in 20 households owns a piano (5%), giving 50,000 pianos. Each piano gets tuned roughly once a year, taking a tuner about 2 hours including travel. A tuner works, say, 1,500 hours a year (about 40 weeks at 37.5 hours). So the number of tuners needed is:

$$\frac{50{,}000 \times 2}{1{,}500} \approx 67 .$$

The real number is somewhere in the dozens, so an estimate landing near 67 is exactly the kind of ballpark accuracy the question is designed to reward — being off by 2x is fine, being off by 100x means a broken assumption somewhere in the chain.

What this means in practice

These questions are common in quant and trading interviews specifically because a desk needs people who can put a rough, defensible number on something unknown under time pressure — sizing a new market, estimating a competitor's flow, or gauging how much liquidity a new product might attract. The grading criterion is the reasoning chain, not the final digit: state each assumption explicitly, keep every factor round and defensible, and sanity-check the final order of magnitude before committing to it.