Market sizing questions: the method, and three worked examples

How do you answer a market sizing question?

A market sizing question asks you to estimate a number, usually a market's annual value or volume, from assumptions you state out loud. Pick top-down or bottom-up, name it, break the estimate into three or four steps you can multiply in round numbers, compute each stage audibly, then sense-check the result against a per-capita implication and give it with a range.

What to take away

  • Interviewers do not compare your answer to a real figure: they mark whether the chain is auditable and the assumptions are stated
  • A defensible answer 30% off the truth outscores a lucky exact number reached in silence
  • Say whether you are going top-down or bottom-up before you start multiplying, then stay in that frame
  • Round every assumption to a number you can multiply in your head, and say that you have rounded
  • Sense-check by converting the result back into a per-person or per-unit implication before you quote it
  • Always state the final number with a range: an estimate you never land is unscoreable

What is a market sizing question actually testing?

A market sizing question sounds like a maths problem and is marked like a communication problem. The interviewer sometimes has a rough figure in mind and often does not. What they always have is a scorecard, and the cell your estimate lands in is rigor: the sub-skill that marks quantitative work inside the three dimensions and fourteen skills we mark against. Rigor is defined as maths that is right, or right in order of magnitude, with assumptions said out loud and a sense-check at the end. Two of those three clauses are about what you say, not what you compute.

This is why the folklore question, "what is the right number", is a category error. Nobody in the room knows the annual revenue of coffee shops in Manchester. The interviewer knows what a good estimate looks like: three or four steps, each a round number they can follow, arriving somewhere plausible. A candidate who lands within 30% of a defensible figure while narrating every assumption outscores one who hits the exact number after ninety seconds of silence, because the second candidate produced nothing to mark.

The second thing on trial is presence, and candidates underrate it here. Market sizing is the one part of a case where you do arithmetic out loud while somebody watches. Multiplying 400,000 by 3.5 is trivial on paper and much less trivial when you are also managing your voice, a blank sheet, and the knowledge that a wrong digit will be heard. Presence marks staying calm under that and thinking aloud without losing the thread. Underneath both is a quieter test: whether you will commit to a number about something unknowable, which is what firms sell.

  • Rigor: is the chain auditable, are the assumptions stated, is there a sense-check at the end
  • Presence: can you compute out loud, calmly, without narrating your own anxiety
  • Commercial judgement: does the final number imply something plausible about real people spending real money
  • Precision: did you actually state the number, in the units you were asked for

What is the method for answering a market sizing question?

Every estimation question yields to the same five moves. The order matters more than the arithmetic, because four of the five are things you say rather than things you calculate.

01

Clarify what is being sized, and in what units

Ask two questions before you start: what exactly counts, and do they want value or volume. "The market for coffee" could mean cups sold, retail revenue, or beans imported, and those differ by orders of magnitude. Confirm the geography, the period (almost always a year) and the currency. Thirty seconds here prevents a five-minute answer to the wrong question.

02

Choose top-down or bottom-up, and say which

Top-down starts from a large total and cuts it down with percentages. Bottom-up starts from a single unit of demand, one customer or one car, and multiplies up. Say the words: "I'll build this bottom-up from an individual consumer." Naming the approach tells the interviewer how to follow you, and it stops you drifting between the two halfway through, which is where most chains break.

03

State assumptions in round numbers you can multiply

Use 60 million rather than 58.4 million, £3.50 rather than £3.42, one car per three people rather than 0.36 cars per person. Label each one plainly: "I'll assume, as a round figure, that ...". Rounding is not sloppiness, it is a stated modelling choice, and an interviewer will hold you to what you said rather than to reality.

04

Compute in named stages, out loud

Do not run the whole multiplication in one breath. Announce a stage, compute it, state the intermediate result, then move on: population, then buyers, then daily spend, then annual spend. Named stages give the interviewer somewhere to intervene and give you somewhere to restart if you fumble a digit. Write each intermediate down with its unit attached.

05

Sense-check, then give the number with a range

Convert the answer back into something human before you quote it: spend per person per year, or units per household. If the implication is absurd, say so and name the assumption you would revisit. Then commit: "About £400 million a year, and I'd put a range of £260 to £420 million on it, driven mainly by the participation assumption."

The choice in step two is not arbitrary. Bottom-up is stronger when the unit of demand is a person doing something repeatedly: buying coffee, replacing tyres, renewing a subscription. Top-down is stronger when there is a known total to carve, such as a population, a vehicle fleet, or a physical volume. If both routes are open, doing one properly and naming the other as a cross-check is the strongest available move.

Worked example: how do you size coffee shop revenue bottom-up?

The question: estimate the annual revenue of coffee shops in a city of one million people. This is a bottom-up build, because the unit of demand is one person buying one coffee.

Coffee shop revenue in a city of 1 million
Approach: bottom-up, from one consumer on one day up to annual city revenue. Every figure below is a round assumption I am choosing, not researched data.
Population: 1,000,000 people.
Assume 40% of them buy a coffee out on a given weekday. That is 400,000 buyers.
Assume one cup each, so 400,000 cups a day.
Assume an average ticket of £3.50, blending £2.50 filter coffee with £4.50 speciality drinks.
Daily revenue: 400,000 × £3.50 = £1,400,000. Call it £1.4 million a day.
Assume 300 trading-day-equivalents a year, because weekends and holidays run lighter than weekdays across 365 calendar days.
Annual revenue: £1.4 million × 300 = £420 million.
Sense-check: £420 million across 1,000,000 people is £420 a head per year. Per buyer it is £1,050, which is 300 cups at £3.50. That is a committed daily habit for 40% of a city, so 40% is the assumption I would flex first.
At 25% participation instead: 250,000 cups × £3.50 = £875,000 a day, and £875,000 × 300 = £262.5 million.
The answer: roughly £400 million a year, with a range of £260 million to £420 million. Participation is the single biggest lever.

Three things in that chain earn marks independently of whether £420 million is close to the truth. The average ticket was justified rather than asserted, by naming a cheap drink and an expensive one and blending them. The 300 trading-day-equivalents was explained, so the interviewer knows it is a deliberate haircut on 365 rather than a number from nowhere. And the sense-check found something real: £1,050 per buyer per year is high, the candidate said so unprompted, then quantified what a more conservative assumption does to the answer.

Notice also what the candidate did not do: switch mid-answer to a second chain built from the number of coffee shops times revenue per shop. That route makes a fine cross-check, but blending two chains is how estimates double-count. Finish one, then offer the other.

Worked example: how do you size a market top-down with segmentation?

The question: estimate the annual market for replacement car tyres in a country of 60 million people. This one goes top-down, because there is a total worth carving: the national vehicle fleet.

Replacement car tyres in a country of 60 million
Approach: top-down from population to vehicle fleet, then segmented by how hard each segment drives. Every figure below is a round assumption, not a researched statistic.
Population: 60,000,000.
Assume one car per three people, which folds in children and non-drivers. Fleet: 60,000,000 ÷ 3 = 20,000,000 cars.
Assume four tyres per car, replaced as a set every four years. That averages one tyre per car per year.
Flat volume: 20,000,000 cars × 1 tyre = 20,000,000 tyres a year.
At an assumed £70 per tyre fitted: 20,000,000 × £70 = £1,400,000,000, so £1.4 billion a year.
Now segment, because fleets do not drive alike. Assume 80% private cars (16,000,000) and 20% commercial or fleet vehicles (4,000,000).
Private: tyres last the assumed four years, so 1 tyre per car per year, giving 16,000,000 tyres.
Commercial: higher mileage, so assume tyres last two years, giving 2 per vehicle per year. 4,000,000 × 2 = 8,000,000 tyres.
Segmented volume: 16,000,000 + 8,000,000 = 24,000,000 tyres a year.
Price the segments separately. Private: 16,000,000 × £70 = £1.12 billion. Commercial: 8,000,000 × £90 = £720 million.
Segmented value: £1.12 billion + £0.72 billion = £1.84 billion.
Sense-check: 24,000,000 tyres across 20,000,000 cars is 1.2 tyres per car per year, and £1.84 billion across the fleet is about £92 per car per year. Just over one tyre a year per vehicle reads right for a mixed fleet.
The answer: about £1.8 billion a year, with a band of £1.4 billion to £2 billion. The replacement interval moves it most: stretching private replacement from four years to five takes that segment from 16,000,000 tyres to 12,800,000, which is about £220 million off the total.

The segmentation is what separates a 3 from a 4 here. The flat chain reaches £1.4 billion in four moves and is entirely defensible. The segmented version does something extra: it notices that "a car" is not one thing, splits on the variable that actually drives replacement, which is mileage, and prices the two segments differently. That is commercial judgement showing up inside a maths answer.

Segment on one variable, though, not three. Private against commercial is worth the effort because it changes the replacement rate materially. Splitting further by hatchback, saloon and estate adds three sub-calculations and no insight, and it eats the clock you need for the rest of the case. If a cut does not change a rate or a price, do not make it.

Worked example: how many golf balls fit in a Boeing 747?

Physical-volume brainteasers are the purest form of the exercise, because nobody pretends there is a research report to check against. The golf-balls-in-a-747 question is the classic, and it is the one we use for the free three-minute spoken taster. It resolves into three quantities: the volume of the container, the volume of one ball, and a haircut for the space you cannot fill.

Golf balls in a Boeing 747
Approach: cabin volume divided by ball volume, then a haircut for fittings. Every dimension below is a round approximation I am stating up front.
Approximate the cabin as one cylinder: assume a radius of 3 metres and a usable length of 60 metres. I am ignoring the upper deck, the cargo hold and the taper at nose and tail, and treating the whole thing as a clean tube.
Cabin volume: π × r² × length, taking π as 3. So 3 × 3² × 60 = 3 × 9 × 60 = 1,620 cubic metres. Call it 1,600 cubic metres.
Approximate one golf ball by its bounding cube, 4 centimetres on a side: 4 × 4 × 4 = 64 cubic centimetres per ball.
Convert the cabin into the same units. One cubic metre is 100 × 100 × 100 = 1,000,000 cubic centimetres, so 1,600 cubic metres is 1,600,000,000 cubic centimetres.
Balls before any haircut: 1,600,000,000 ÷ 64 = 25,000,000. Twenty-five million.
Cross-check by the other route: one cubic metre holds 1,000,000 ÷ 64 = about 15,600 balls, and 1,600 × 15,600 is about 25 million. The two routes agree.
The bounding cube has already handled sphere packing: a ball fills only about half of its own cube, and that void is baked into the 64. So the remaining haircut is for objects inside the cabin, not for gaps between balls.
Haircut for seats, galleys, lavatories and overhead structure: assume they take 20% of the volume. 25,000,000 × 0.8 = 20,000,000.
The answer: about 20 million golf balls, which is to say the order of magnitude is ten million rather than one million or a hundred million. I would not defend the second digit.

The point of that answer is not 20 million. Change the cabin radius from 3 metres to 2.5 metres and the volume falls by about 30%, taking the count to roughly 14 million. Both answers are correct, because both are the honest output of stated assumptions. What would be incorrect is an unstated assumption, a conversion done in silence, or a final answer in the thousands, which tells the interviewer the candidate mislaid three decimal places and did not notice.

Unit conversion is where this question actually kills people. Metres to centimetres is a factor of 100, but cubic metres to cubic centimetres is a factor of a million, because the cube applies to the conversion too. Announce the conversion as its own stage and the error becomes hard to make.

To rehearse the spoken version rather than the written one, the taster at /try runs an estimation question as a three-minute conversation with no signup, and the timed estimation reps in /drills exist because doing this against a clock is a different skill from doing it on paper.

How do you sense-check an estimate, and what goes wrong most?

The sense-check is the cheapest mark available and the most commonly skipped. It takes one sentence and it works the same way every time: divide your answer by the population, the household count or the unit count, and ask whether the result describes a person or an object you can picture.

£420 a head per year on coffee out. £92 per car per year on tyres. Twenty million golf balls in an aircraft that carries around 400 passengers, which is 50,000 balls per seat. Each of those is a sentence you can say in four seconds, and each either confirms the chain or catches an order-of-magnitude error before the interviewer has to.

Say the number, then the range

Ending an estimate without a number is the one unrecoverable error, and it is more common than it should be. Uncertain candidates trail off into "so it would be somewhere in that region", and the rigor cell has nothing in it. Commit to a figure, attach a band, then name the assumption driving the band. A stated range reads as judgement rather than doubt: it shows you know which of your assumptions is load-bearing.

The five errors that cost most

ErrorWhat it looks likeThe fix
Base off by an order of magnitudeSizing a country of "6 million" when you meant 60 million, or treating a mid-sized city as 10 million peopleState the base out loud as a round assumption and pause on it for one beat. A wrong base makes every later step worthless, so it is the one number worth double-checking
Unit driftFolding a monthly price into an annual volume, or dividing cubic metres by cubic centimetres without convertingWrite the unit next to every intermediate result, and make each conversion its own announced stage instead of hiding it inside a multiplication
An implausible per-capita implicationAn answer that quietly implies every adult buys nine mattresses a year, left unexaminedDivide the final answer by the population before you quote it and say what the per-person figure is. If it is absurd, name the assumption you would change
Silent arithmeticNinety seconds of quiet, then "I get about four hundred million"Announce each stage and its result as you go. An audible chain that ends wrong marks far better than a silent one that ends right
Never stating the final number"So it's a large market, in the hundreds of millions, depending on assumptions"Give one figure, then the range, then the biggest lever. Hedging is the most expensive habit in the whole exercise
Five failure modes, ordered by how much ground they lose. Four of the five are process errors rather than arithmetic ones.

Common questions

What is a market sizing question in a case interview?

It is a request to estimate a quantity nobody in the room knows precisely: the annual value of a market, the number of units sold in a country, or the volume of a physical container. You are given almost no data, on purpose. The question exists to show how you build and defend a numerical estimate from stated assumptions, not to test what you have memorised.

How accurate does a market sizing answer need to be?

Close enough to be the right order of magnitude, which in practice means within roughly a factor of two or three of a defensible figure. Interviewers are not holding a true number against your answer, because usually there is not one. What does lose marks is being out by a factor of a thousand, since that signals a unit or base error nobody in the chain caught.

Should I use top-down or bottom-up for market sizing?

Use bottom-up when the unit of demand is a person doing something repeatedly, such as buying coffee or replacing tyres. Use top-down when there is a known total you can carve, such as a national population, a vehicle fleet or a physical volume. Either is acceptable. What matters is naming which one you are using before you start, and not switching halfway through.

Is it acceptable to make up population figures in a market sizing question?

Yes, provided you label them as round assumptions rather than facts. Saying you will assume a country of 60 million, or a city of one million, is exactly what an experienced consultant does at the start of an estimate. Offering a precise-sounding figure you are not sure of is worse, because it implies an accuracy your chain cannot support and invites a correction.

How long should a market sizing answer take?

Two to four minutes of speaking for a standalone estimation question, including the twenty or thirty seconds you take to plan before you talk. Inside a longer case, a sizing step is usually shorter still. If you are past five minutes you are almost certainly over-segmenting, so cut the number of branches rather than the sense-check at the end.

How do I practise market sizing questions out loud?

Set a three-minute timer, pick a question, and speak the whole chain including every assumption and the closing sense-check, ideally recording yourself. Playing the recording back shows you where the arithmetic went silent, which is the failure you cannot hear in the moment. Our free spoken taster runs one estimation question with no signup, and the drill set includes timed estimation reps.