Breaking an Estimate Into a Decomposition Tree

Estimate how many gallons of gasoline are consumed by cars in the United States in one year.

Try to get a number in the next two minutes before reading on. If your instinct is "I have no idea," that's exactly the trap — nobody knows this number from memory, but almost everyone can build it.

Why one guess is worse than several

A single blind guess at "gallons of gas per year" could be off by a factor of 100 in either direction — you have no anchor to check it against. But if you instead guess the number of cars, the miles each drives per year, and the miles per gallon each gets, each of those sub-guesses is something you have real intuition about, because you've seen a car, driven somewhere, or filled a tank. The trick is to write the target as a product of factors you're each individually confident within a factor of 2 of, rather than one factor you're confident within a factor of 100 of.

A decomposition tree is just that written down as a tree: the target quantity at the root, and at each level you split it into a product (or sum) of pieces, stopping when every leaf is something you can look up or genuinely estimate from lived experience.

Solving the gasoline estimate

Leaf 1 — number of cars. The US population is about 330 million. Roughly 1.7 people per household car on average (some households have none, some have three), so call it 1 car per 1.7 people: $330\text{M} / 1.7 \approx 190\text{ million cars}$.

Leaf 2 — miles driven per car per year. A typical commute-plus-errands driver covers roughly 12,000 miles a year — about 230 miles a week, which matches a daily commute plus weekend trips.

Leaf 3 — fuel efficiency. A typical car gets around 25 miles per gallon, so it burns $1/25$ gallons per mile.

Multiply the leaves:

$190{,}000{,}000 \text{ cars} \times 12{,}000 \frac{\text{mi}}{\text{yr}} \times \frac{1}{25}\frac{\text{gal}}{\text{mi}} = 91{,}200{,}000{,}000 \text{ gallons/yr}$

Answer: roughly 90 billion gallons per year. The real US figure (EIA data) is around 130–140 billion gallons for all motor gasoline, so this back-of-envelope tree lands within a factor of about 1.5 — well inside the "right order of magnitude, plausible first digit" bar an interviewer is actually checking.

A second example: piano tuners in a city

How many piano tuners are there in a city of 3 million people? Decompose: (pianos per household) × (households) gives total pianos; (tunings per piano per year) tells you total tuning-jobs per year; (jobs one tuner can do per year) converts that into the number of tuners needed.

• Households: 3,000,000 people / 2.5 per household ≈ 1.2 million households.
• Pianos: maybe 1 in 20 households owns one → 60,000 pianos.
• Tunings needed: once a year each → 60,000 tuning-jobs/year.
• One tuner's capacity: say 4 tunings a day, 5 days a week, 48 weeks a year → 960 tunings/year per tuner.
• Tuners needed: $60{,}000 / 960 \approx \textbf{63 piano tuners}$.

Each leaf here is a number an ordinary adult can defend in one sentence — that's the test for whether the tree is decomposed finely enough.

The generalizable skill

The decomposition tree is the reusable move behind every market-sizing and Fermi question a quant interview throws at you: total notional traded in an asset class, a firm's annual revenue, an exchange's daily volume. The skill being tested isn't arithmetic — it's structuring an unknown as a product of knowns, choosing sensible leaves, and being able to defend each branch out loud. Once the tree is built, the multiplication is the easy part.