How would you estimate the number of piano tuners working in Chicago from scratch?
answer
- chain of factors, not one guess
- start from city population
- households, then piano ownership rate
- tunings a tuner fills in a year
- divide annual demand by annual capacity
basics
~20 sBreak the target number into a chain of estimable factors: city population, people per household, share of households owning a piano, tunings per piano per year, and tunings one tuner performs per year. Multiply through, divide, and land near 50.
solid answer
~50 sI would decompose the count into a chain of factors I can each guess within a factor of two or three, then multiply. Chicago has roughly 3 million residents; at about 2.5 people per household that is around 1.2 million households. If about one household in 25 owns a piano, that is roughly 50,000 pianos, and institutions such as schools and churches add a modest amount on top. A piano that is actually played gets tuned about once a year, so demand is on the order of 50,000 tunings per year. On the supply side, a tuner doing 4 tunings a day, 5 days a week, 50 weeks a year handles about 1,000 tunings a year. Dividing 50,000 by 1,000 gives roughly 50 tuners. I would report the answer as tens rather than hundreds, and name the assumption I am least sure of.
go deeper
Be ready to say the chain of factors out loud before touching arithmetic, and to keep every number round. Reciting a decomposition confidently matters far more here than landing on any particular final figure.
An interviewer at this level expects you to explain why multiplying five rough guesses still yields a usable answer, and to rerun the whole chain in one sentence when a factor is challenged mid-answer.
Show that you finish with a range and a named weakest assumption rather than a single number, and that you sanity-check the result against a density you already know before presenting it.
Own the question of what resolution the answer actually needs. Argue when an order of magnitude is enough to settle the decision at hand and when the chain has to be tightened, so nobody spends a week refining a factor that cannot change the conclusion.
## What is being asked A question like *how many piano tuners work in Chicago* is a **Fermi problem** — named after the physicist Enrico Fermi, who was known for producing usable estimates of quantities nobody had measured. The interviewer is not testing whether you know the answer. They are testing whether, faced with a number you cannot look up, you can build a defensible chain of reasoning out loud instead of freezing or guessing. The method has one core move: **replace one unknowable number with several guessable ones**. You cannot guess the tuner count directly, but you can guess a city's population, roughly how many people live in a household, and roughly how many tunings fill a working year — because each of those is anchored to everyday experience. ## The decomposition A clean version runs in two halves that meet in the middle: estimate **demand** (tunings needed per year), estimate **supply capacity** (tunings one tuner delivers per year), and divide. **Demand side** 1. Chicago's population: about 3 million in the city proper. (If you use the wider metro area, say so — that is a different question and a bigger answer.) 2. People per household: about 2.5, so roughly 1.2 million households. 3. Share of households with a piano: call it 1 in 25, or 4%. That gives about 48,000 — round to 50,000 pianos. 4. Institutional pianos (schools, churches, bars, music venues) add some more; they are a modest fraction, not a multiplier, so noting them and moving on is fine. 5. Tunings per piano per year: a piano that is regularly played is typically tuned about once a year; many household pianos are barely played and go years untuned. Take 1 per year as the working figure. Demand is therefore on the order of **50,000 tunings per year**. **Supply side** 6. A tuner does perhaps 4 tunings in a working day, allowing for travel between jobs. 7. Working 5 days a week, 50 weeks a year: 4 x 5 x 50 = **1,000 tunings per tuner per year**. **Divide**: 50,000 / 1,000 = **about 50 piano tuners**. ## Why the errors do not blow up A natural objection: five guesses, each possibly wrong, must compound into nonsense. In the worst case they do — five factors each off by 2x in the same direction is 2^5 = 32x. But the guesses are not systematically biased in one direction. Some run high, some run low, and in a product they partially cancel. That is why Fermi estimates typically land within a factor of a few, which is exactly the resolution these questions want. The right way to state the result is **tens of tuners, not hundreds and not thousands**. This also explains why precision in any single step is wasted effort. Using 2.54 people per household instead of 2.5 does nothing when the piano ownership rate could be 2% or 8%. ## What the interviewer is grading - **Structure before arithmetic.** Say the chain of factors out loud before you start multiplying, so the interviewer can challenge a factor rather than an answer. - **Round numbers.** Work in 3 million, 50,000, 1,000. Carrying four significant figures through a chain of guesses signals that you have confused a guess with a measurement. - **Named assumptions.** Every factor should be spoken as an assumption you are willing to revise, not smuggled in silently. - **A stated confidence.** Finish with a range and a direction: about 50, plausibly 25 to 100, and here is the number that would move it most. - **Recovery under challenge.** If the interviewer says the ownership rate is closer to 1 in 10, you should be able to rerun the chain in one sentence: pianos go to about 120,000, tuners to about 120. ## Common failure modes **Guessing a single number.** Answering *maybe 200* with no chain gives the interviewer nothing to evaluate. **Refusing to estimate.** Saying you would look up the data is the answer that fails hardest — the whole point is what you do before the data exists. **Confusing population with households.** Multiplying 3 million people by a per-household ownership rate overstates pianos by roughly the household size. **Forgetting to convert demand into tuner-years.** Stopping at 50,000 tunings answers a different question. The final division by one tuner's annual capacity is what turns a volume of work into a headcount. **Silent unit drift.** If one factor is per week and another per year, the product is meaningless. Say the unit of every factor aloud. ## A useful habit After the multiplication, do a rough plausibility check against something you already know: 50 tuners in a city of 3 million is one tuner per 60,000 people, which is roughly the density you would expect for a rare specialist trade — rarer than dentists, comparable to a niche craft. If your arithmetic had produced 50,000 tuners, that check would have caught it instantly.
- Suppose the true count turns out to be four times your estimate. Did your estimate fail?Not in the sense the exercise cares about. A Fermi estimate is judged on order of magnitude and on whether the reasoning is auditable. Landing at 50 when the truth is 200 means the chain was sound and one factor was low — most likely the piano ownership rate or tunings per piano. Being off by 4x is a normal result; being off by 1000x means a factor is missing or a unit slipped.
- How do institutional pianos change the decomposition?They add a second demand term rather than scaling the first. Schools, churches and venues own far fewer pianos than households do, but each is tuned more often — several times a year rather than once. Estimate them as their own small chain and add it to the household total. If the added term is under about 20% of the household figure, it does not move the order of magnitude and can be acknowledged and dropped.
- You are unsure whether Chicago has 3 million or 9 million people. How do you proceed?State the ambiguity and pick one, because the two numbers answer different questions: 3 million is the city, roughly 9 million is the metro area. Choose the city, carry it through, and note that using the metro figure multiplies the answer by about three. Making the choice explicit is worth more than getting it right silently, because the interviewer can then redirect you in one sentence.
It is like guessing the weight of a loaded truck: you cannot lift it, but you can guess the weight of one crate, the crates per pallet, and the pallets on board, then multiply.
saying these in an interview costs you the question
- Guesses a single number with no decomposition at all
- Refuses to answer without looking up real data
- Carries four significant figures through a chain of guesses
- Multiplies population by a per-household ownership rate
- Stops at total tunings and never divides by tuner capacity
- Mixes weekly and annual factors in the same product