Fermi Estimation: How to Guess Any Number in Sixty Seconds
A Fermi problem asks how many piano tuners live in Chicago, and a decent estimator lands within a factor of three in under sixty seconds without a calculator. The trick is a chain of five order-of-magnitude guesses, each rounded to a power of ten, multiplied together, then audited for cancellations. Enrico Fermi named the technique in 1945 when he dropped paper scraps from the Trinity blast tower and estimated a yield near ten kilotons, inside a factor of two of the measured value that took Los Alamos three weeks to confirm. The same routine sizes markets, forecasts server bills, and cuts through product-meeting arguments about traffic, and the arithmetic stays in your head.
Why Order-of-Magnitude Beats a Spreadsheet
Most inputs to a real question are known to a factor of two at best. An exact-looking spreadsheet answer is a lie about that uncertainty, because one wrong input silently corrupts every downstream cell. A Fermi estimate rounds every input to the nearest power of ten before multiplying, and the errors cancel roughly as often as they compound. A five-factor chain with each input off by a factor of three still lands the total inside a factor of ten. Consulting firms teach the method because it fails visibly instead of silently, and because a partner can audit a five-line estimate in thirty seconds. The same math powered the Trinity paper-drop in 1945, the original Drake equation for detectable civilizations in 1961, and the sizing question in almost every product-strategy interview.
The Five-Step Chain
Every Fermi problem decomposes into a product of five to seven factors, each easier to guess than the whole. State the target as a rate times a stock times a share, or as a probability times a population, and each factor becomes a one-order guess. Round every factor to 10^k, keep the exponent on a mental register, and add exponents instead of multiplying values. Bring the leading digits back at the end with one multiplication of numbers between 1 and 10. Sanity-check the answer against a known anchor before publishing it.
- Rewrite the target quantity as a product of factors known to an order of magnitude.
- Round each factor to the nearest power of ten and record only its exponent.
- Sum the exponents to get the answer's order of magnitude.
- Multiply the leading digits (values between 1 and 10) in a separate second pass.
- Compare the result against one real-world anchor and flag any factor-of-ten mismatch.
Piano Tuners in Chicago in Sixty Seconds
Chicago holds about 3 million people, or 3 x 10^6. Roughly one household in twenty owns a piano, and average household size sits near 2.5, so piano count is 3 x 10^6 divided by 2.5 divided by 20, which rounds to 6 x 10^4. A piano needs a tuning once a year on average, giving 6 x 10^4 tunings per year. A working tuner handles four pianos a day for 200 working days, or 800 tunings a year, so the city supports 6 x 10^4 divided by 800, near 75 tuners. The 2020 U.S. Bureau of Labor Statistics regional file listed 78 piano tuners in the Chicago metro area, which is inside the estimate by less than 5 percent. Not every problem lands that close, but the chain is transparent enough that the wrong factor is obvious in retrospect.
Fermi at the Grocery Store, the Poker Table, and the Product Meeting
A Fermi chain sizes weekly grocery spend as items per trip times price per item times trips per month, and any receipt over 50 percent of that estimate is worth checking. At the poker table, chip stacks times average pot times hands per hour times rake fraction gives the room's take, which prices the seat before you sit down. Product meetings collapse when someone asks how many daily users a feature needs to justify one engineer, and a five-factor chain answers in under two minutes. The same skill turns percentage shortcuts and long-column addition into raw inputs for larger estimates. A few rounds at /menu lock the digit-shifting reflex the method leans on.
The Errors That Kill an Estimate
The most common failure is double counting, where two factors quietly measure the same quantity, so the product overshoots by a full order of magnitude. Missing a unit conversion between hours and years, or between dollars and thousands, is the second. Anchoring on a memorable number and letting later factors bend around it, named as anchoring bias in the 1974 Tversky and Kahneman paper, corrupts the chain silently. Rounding all factors up, or all down, instead of alternating pushes the answer off by a factor of five inside a five-step chain. A final anchor-check catches all four errors at once, which is why professional estimators refuse to skip it.
- Double counting the same quantity in two different factors.
- Missing a unit conversion between time, money, or population scale.
- Anchoring on a memorable input and bending later factors to fit it.
- Systematic rounding in one direction across every factor.
- Skipping the final real-world anchor check.
A Ten-Day Fermi Drill
Run one estimate a day for ten days, timed at ninety seconds each, and keep a two-column log with your answer and the true value. Days 1 to 3 are population questions like tuners, pizza shops, and dogs in a target city. Days 4 to 6 shift to money: monthly coffee spend, annual laundromat revenue, city payroll. Days 7 to 9 turn to rate questions: heartbeats in a lifetime, breaths per year, downloads on the top mobile app. Day 10 is a hard chain of seven factors, and the finish line is landing inside a factor of three on eight of ten problems. The six-week mental-math rebuild plan folds this drill into week three for readers who need supporting arithmetic reps.
