A five-step method
- Clarify exactly what is being estimated, and the scope (country, time period, units).
- Choose an approach: top-down from population, or bottom-up from capacity.
- Break it into factors you can estimate, and state each assumption out loud.
- Calculate with round numbers, writing them down.
- Sanity-check against something you know, and give a range.
Worked example: gas stations in the US
Top-down via demand. About 330 million people and roughly 280 million registered vehicles; call it 250 million cars in regular use. A car fills up about once a week: 250 million fill-ups a week, around 35 million a day. A station with eight pumps, each serving maybe four or five cars an hour over sixteen busy hours, handles very roughly 500 to 600 fill-ups a day once you account for quiet periods. 35 million ÷ 500 ≈ 70,000; at 250 a day it would be 140,000. A range of 70,000 to 150,000 is defensible — published estimates are around 145,000, which shows why stating a range and the sensitive assumption (fill-ups per station) matters.
Worked example: storage for a photo app
Assume 50 million daily active users, 10% upload on a given day, two photos each: 10 million photos a day. At about 3 MB per stored photo including a compressed version and thumbnails: 30 TB a day, roughly 11 PB a year before replication. With three replicas, about 33 PB a year. This is the kind of estimate a system design interview expects in the first five minutes, and the obvious next point is whether to tier old photos to cheaper storage.
Where estimation questions show up
| Interview | Typical question | What's assessed |
|---|---|---|
| Consulting case | Market size for electric scooters in a city | Structure, business sense |
| Product management | How many people use a feature daily? | User reasoning, metrics sense |
| System design | Requests per second, storage, bandwidth | Capacity planning |
| Quant trading | Quick numerical estimates under pressure | Speed and calibration |
Top-down or bottom-up?
Top-down estimates start from a large known number — population, households, vehicles — and narrow it with percentages. Bottom-up estimates start from a unit you can picture — one store, one server, one delivery driver — and multiply out by capacity. The gas station example above is really both: demand from the top, capacity per station from the bottom. Doing the problem both ways, even roughly, is the best sanity check you can offer. If the two answers are within a factor of two or three, say so with confidence; if they are far apart, name the assumption most likely to be wrong. Interviewers in consulting particularly like hearing a candidate reconcile two approaches.
Arithmetic shortcuts
- Round aggressively: 330 million becomes 300 million unless precision matters.
- Use powers of ten: 3 × 10⁸ × 2 × 10⁻¹ = 6 × 10⁷.
- Remember a day has about 86,400 seconds — call it 100,000 for quick requests-per-second estimates.
- A year has about 31.5 million seconds; 365 days is roughly 400 for quick division.
- Write numbers down; mental slips are the most common failure.
Mistakes to avoid
Guessing a single number without decomposition, silently assuming, getting lost in precision, and never sanity-checking. Another is freezing because you don't know a base fact such as a city's population — say your best guess and move on; the interviewer cares about the structure. If the interviewer offers a different assumption, adopt it and recalculate cheerfully.
Practising
Do two estimation questions a day for two weeks, out loud, timing each at five to eight minutes. Check your answers against published figures afterwards to calibrate. A free AI interview assistant such as CoPilot Interview can run estimation questions in a mock interview and question your assumptions. Our quant interview prep guide covers faster-paced estimation.
Disclosure: CoPilot Interview is our product, so weigh what we say about it accordingly. Everything else on this page applies whichever tool you use, or none.
Frequently asked questions
What is a Fermi estimation question?
A question asking you to estimate a quantity you can't look up — how many piano tuners work in a city — by breaking it into factors you can reasonably estimate and combining them.
How accurate do estimation answers need to be?
Within an order of magnitude is usually fine. Interviewers care more about clear decomposition, stated assumptions, clean arithmetic and a sanity check than a precise number.
How do estimation questions appear in system design interviews?
As back-of-the-envelope capacity estimates — requests per second, storage per year and bandwidth — that guide design choices such as caching and sharding.
What if I don't know a base number like population?
State a reasonable guess and continue. The interviewer is evaluating the method, and may correct the number for you.
Practise on real interview questions
Run 145 real interview questions across 10 tracks, answer out loud and see a structured answer to compare with yours. Free, no signup.
Try a practice question