Named for the physicist Enrico Fermi, whose order-of-magnitude estimates were known from the 1940s.
Estimates an unknown quantity within an order of magnitude by breaking it into a chain of factors, guessing each, and multiplying them.
How many piano tuners a city of a million people supports
Each slider is one factor in the chain, and the answer is their product: people become households, households become pianos, pianos become tunings, tunings become tuners. Change a guess and watch it run through. The bounds multiply every factor at its worst; the piano share is the widest guess, so narrowing it narrows the range most. Illustrative.
An example
A two-person startup building booking software for piano technicians had a week to choose a launch city and no industry data worth the name. The founder sized one metro area of about a million people at a whiteboard in ten minutes. At about 2.5 people to a household, a million people is 400,000 households. If one household in twenty owns a piano, that is 20,000 pianos, and at roughly one tuning a year, 20,000 tunings. A tuner doing four a day over 250 working days handles about 1,000 a year, so the city supports about 20 tuners.
Then she bounded it. Piano ownership was the shakiest guess, anywhere from one household in fifty to one in eight, and on its own it swung the answer from 8 to 50. She rebuilt the count a second way, from a regional directory of tuners, and got 26. Twenty-odd customers a city was too few to pay for a sales team, so the startup chose to sell through a national trade association.
Why it works
Nobody knows how many piano tuners a city holds, but most people can reason about household size, how common pianos are, and how many jobs a tradesperson can fit into a day. Breaking the unknown into factors like these turns one wild guess into several tame ones, each checkable against ordinary experience.
The errors also tend to cancel. When each guess is independent, some run high and some run low, so the product lands closer to the truth than any single factor's error would suggest. An answer within a factor of two or three is enough for many decisions, such as whether a market can carry a sales team. And the chain shows where the uncertainty lives: the factor with the widest range is the one worth an hour of research.
The thinking behind it
Enrico Fermi Enrico Fermi, the Italian-American physicist, was known for quick, approximate answers drawn from very little data. At the Trinity test in July 1945 he dropped scraps of paper into the blast wave and, from how far they moved, estimated a yield of about ten thousand tons of TNT. That was roughly half the measured figure and well within an order of magnitude. He set down the account in My Observations During the Explosion at Trinity on July 16, 1945. Problems answered by chaining rough factors in this way now carry his name.
How to run it
- State the quantity and its unit. Write the number needed in one line, with its unit and its scope: piano tuners working in one metro area of a million people.
- Break it into factors that multiply. Pick three to five factors that ordinary experience can judge, chained so their units cancel.
- Guess each factor three ways. Give a low, a best, and a high value, and mark the guess you trust least.
- Multiply the best guesses. Round as you go; one or two significant figures are plenty.
- Bound the answer. Multiply the lows, then the highs, and quote the range beside the best guess.
- Rebuild it another way. Start from a different quantity, such as directory listings or industry revenue, and see whether the two land within a factor of three.
- Tighten the widest factor. If the decision needs more precision, spend the research on the factor with the widest range, since it moves the answer most.
Where it goes wrong
False precision comes first: an estimate of 21.6 tuners invites trust the method cannot support. Round to one or two figures and always quote the range with it.
The second is a broken chain, where a factor is counted twice or the units do not cancel, such as tunings a month multiplied by pianos a year. Write the unit beside every factor and check that they reduce to the unit in the first line.
The third is guessing toward the answer someone hoped for. When the result lands exactly on the target, rebuild it from a different starting quantity. An estimate that holds up both ways is worth using; one that does not shows where to look next.
When to use something else
- Reference Class Forecasting. When records of similar cases exist and the estimate should start from how often such things happened before.
- Monte Carlo Simulation. When the inputs' ranges matter as much as the answer and the decision needs the odds of landing above a threshold.
- Sensitivity Analysis. When a model already exists and the question is which input moves its answer most.
Where it sits in SOLVE eX
Go to the source
- Order-of-magnitude estimation practice, 1940s