Market Sizing Questions: What the Four Dimensions Score

Sanity check, 20 points, and the one most answers score zero on

It carries the fewest points and it is the only dimension where a miss is total. Scope, chain and arithmetic all award partial credit for a partial attempt. The sanity check does not: an answer that ends on the number scores nothing here, because there is nothing to assess.

A check is one sentence, after the figure, testing it against something known. Take a figure you can look up afterwards. The Census Bureau's advance retail report for August 2026 puts grocery store sales at $610,476 million across the first eight months of the year, which annualizes to roughly $916 billion. So a sizing that lands on $2bn for a grocery category is claiming about 0.2 percent of everything Americans spend in grocery stores, and that is the right order for a narrow category.

Two things make a check count. It names an external reference, a population, a total market, a figure the interviewer can hold. And it runs in the direction of doubt: a check that only confirms is not a check.

The reason this dimension exists at all is that a number without one cannot be distinguished from a guess. An interviewer has no way to tell a disciplined chain from an invented one except by watching whether the candidate tests it.

Assumption chain, 30 points, the heaviest

The chain carries the most weight because it is the part that transfers. Arithmetic is verifiable, scope is a sentence, but the chain is where judgment shows.

It scores on linked, defensible steps rather than quantity. Three steps you can each justify beat six that include one you invented. The failure the rubric looks for is a chain that jumps: population, then a share nobody could estimate, then the answer.

Within a chain, the inputs do not carry equal risk, and knowing which is which is what the dimension rewards:

• Anchors are figures you know. A national population, a household count. Cheap and safe; every chain should start from one.
• Ratios divide an anchor. Share of adults of working age, share of households with a car. Most are defensible within a margin that does not move the conclusion.
• Rates are per-period, replacement cycle, purchase frequency. A rate wrong by 2× moves the answer by 2×, with nothing to absorb it.
• Prices are the final multiplier and the most hand-waved. A unit price guessed at 50 when it is 400 makes every careful step before it irrelevant.

Rates and prices carry the answer. Naming the one that does, out loud, is the single highest-scoring habit on this dimension:

"The number turns on replacement cycle. I assumed seven years, at five it is near 18 million units, at ten closer to 9. Seven is where I would start, and the first thing I would check against data."

That converts the interviewer's best follow-up into something you have already said, and it prevents the failure the rubric cannot score around: a candidate defending a figure they should abandon.

Scope, 25 points, and the cheapest to earn

Scope is one sentence before any arithmetic: what is counted, over what period, in what geography. Units or revenue. New sales or the installed base. Annual or lifetime.

It is worth a quarter of the mark and it takes ten seconds, which makes it the most reliably wasted 25 points in the drill set. The rubric awards it on the presence of an explicit frame. You said what you were including and excluding. Whether the frame matches the one the interviewer had in mind does not enter.

That distinction matters more than it looks. A candidate who states a scope and is corrected loses nothing; the correction arrives before the work. A candidate who leaves it implicit and sizes the wrong thing loses the whole answer, and finds out at the end.

Sometimes the case states the scope for you, and then the job is to repeat it. Oliver Wyman's Poseidon Water Park case asks a candidate to estimate the daily revenue of a competitor's park in Dubai, and publishes no figures whatsoever. What it does publish is a boundary and a warning. "Focus on a peak day, such as a Saturday in April with excellent weather." And: "As a competitor, we do not have access to Poseidon Water Park's internal data." A peak-day figure and an average-day figure differ by a large multiple, and a candidate who quietly sizes the average has answered a question nobody asked.

Arithmetic, 25 points, scored on the working

This dimension does not check whether the answer is right. It checks whether the multiplication was shown.

Round early and round hard. 340 million becomes 350, then divide by 3 and take 100. The precision given up is nothing against the precision the inputs never had, and the calculation stays sayable aloud, which is the real constraint, because an interviewer who loses the thread of the arithmetic stops assessing it.

One discipline worth the four words it costs: round in alternating directions and say so. "Population up, penetration down, so these partly offset." It tells the interviewer you know rounding has a direction.

Four habits carry this dimension, and they are all mechanical.

1. Drop the zeros and put them back at the end. Work in millions, write 12 rather than 12,000,000, and restore the scale once. Every lost order of magnitude in a sizing answer comes from carrying zeros through five multiplications.
2. Multiply in the order that keeps numbers small. Divide before you multiply where you can. Three million households divided by ten before multiplying by four kL is easier than the reverse.
3. Say the running total after every step. It gives the interviewer a place to stop you, which is cheaper than being stopped at the end.
4. Announce every rounding as a rounding. "Roughly 2.5 million" when the chain gave 2.46 million is a case skill. Writing 2.5 with no qualifier is an error, and a reader who redoes the chain will find the gap.

The fourth one is the difference between a defensible estimate and a number that happens to be near the truth. It costs one word. Those four habits are the nearest thing to a market sizing cheat sheet worth carrying, and each one fits on a single line.

A published market sizing case, rebuilt line by line

Oliver Wyman's oil and gas case contains a question the firm itself labels market sizing, and it is the best worked market size example available anywhere, because the firm publishes its data and its total and lets you check the middle.

The published inputs: "Bangalore population: 12 million people (3 million households)", municipal coverage of "20% for low-income households, 50% for medium-income households, and 100% for high-income households", and daily consumption of "1 kL for low-income households, 2 kL for medium-income households, and 4 kL for high-income households". The published answer opens with the figure. "The estimated total water demand of 2,500,000 kL per day indicates a substantial market size."

Twelve million people across three million households is four people per household, which is consistent. Then the chain stalls, because the split of households across the three income bands is not published anywhere on the page. Without it the three lines cannot be computed. The firm's own text pins part of it: the answer says low and medium income households "constitute 90% of the population", which fixes the high-income band at 10 percent and leaves the other two open.

So test the splits. At 60, 30 and 10 percent:

That lands 1.6 percent below the firm's 2,500,000. Try 50, 40 and 10 and the total is 300,000 plus 1,200,000 plus 1,200,000, which is 2,700,000, or 8 percent above. Try 70, 20 and 10 and it is 420,000 plus 600,000 plus 1,200,000, which is 2,220,000, or roughly 11 percent below. A 60, 30, 10 split is the only one of the three that reproduces the published total, and that reconstruction is ours rather than the firm's.

Two things follow. The first is the answer to the question the firm actually wants: its own evaluation criterion rewards the "ability to identify and ask for key information needed, for example water consumption levels or different income groups". The missing split is the case. Naming it is the answer the criterion is written to reward.

The second is the insight the table hands over for free. The high-income tenth of households accounts for 1,200,000 kL of the 2,460,000, which is just under 49 percent. So roughly half the served demand sits in one household in ten, because that band uses four times as much water and is fully connected. That is the sentence a client remembers, and it exists only because the chain was built in bands rather than as one average.

Top-down and bottom-up market sizing, side by side

Top-down narrows from a population: everyone, then the qualifying subset, then the buying subset, then spend. Fast, and the default.

Bottom-up scales from a unit: one store's revenue, stores per city, cities. Slower, and worth it when the unit is concrete and the population filters are not.

A top-down market sizing example, on coffee bought in US shops. Start at 340 million people, rounded up to 350 million. Take 70 percent as adults, which is 245 million. Halve that for coffee drinkers, 122.5 million, and call it 120 million, a rounding worth saying out loud. Say one in three buys from a shop in a given week, so 40 million buyers. At two cups a week and five dollars a cup that is 400 million dollars a week, and across 52 weeks 20.8 billion dollars a year.

A bottom-up market sizing example, same market, built from one shop. One shop sells 400 cups a day at five dollars, so 2,000 dollars a day. Across 360 open days that is 720,000 dollars a year. At 40,000 shops nationally, 28.8 billion dollars. Every figure in both chains is ours, chosen round, and not one of them was looked up.

The two chains land 8 billion dollars apart, 28.8 against 20.8, so the bottom-up answer runs about 38 percent above the top-down one. Two inputs own that gap: the one-in-three weekly purchase share and the 40,000 shop count. The shop count is the one to check, because a count of premises can be looked up and a weekly purchase share cannot.

The signal you chose wrong is a ratio you have no basis for. "Share of consumers who would switch provider in a year" is a placeholder dressed as an estimate, and a chain containing one will score low on the heaviest dimension. When a filter turns into a placeholder, the structure is asking for the other direction.

Running both and comparing is a strong answer rather than a hedge. A gap like the 8 billion above is the most interesting thing available to say: one of the two chains holds an assumption that is simply wrong, and locating it is the skill the dimension exists to measure.

What a market sizing number is for

A market at $2bn growing 3% and a market at $2bn growing 25% are the same figure and opposite recommendations.

Most sizing questions are asked because a decision sits behind them, enter or not, build or buy, price here or there. Ending on the implication rather than the number falls outside this rubric, which assesses the estimate itself. It belongs to synthesis, where the recommendation is tested on the opening sentence alone.

The Bangalore question is an example of this working properly. The sizing exists because a later question asks whether entering clears the client's five-year payback test, and the volume the sizing produces is the input to that test. We finish that arithmetic on return and payback, which is where the 10 percent share assumption meets the client's criterion and fails it.

This is also why the growth rate matters more than the level in most sizing answers, and why a candidate should ask for it. A client deciding whether to build capacity cares about the market three years out. A client deciding whether to buy a competitor cares about the market today. The same estimate serves both, and the sentence that closes the answer is different in each case. Ask which decision the figure feeds before you decide what to say about it.

Practicing the dimension you actually lose

Checking your answer against a real figure teaches little, because you will not learn which of five inputs was off.

Work backwards. Build the chain, identify the input carrying it, and look up only that one. A replacement cycle you put at seven when it is four tells you something transferable about how you estimate durability. A population anchor off by 10% tells you nothing worth carrying forward.

That is the loop: build the chain, name the input that decides it, check that one alone, and note whether you run long or short. Most people are consistent in one direction, and finding yours is worth more than another fifty market sizing practice questions.

Then practice the thing the Bangalore case teaches, which is naming a missing input rather than inventing one. Take any sizing prompt, list the inputs your chain needs, and mark the ones the prompt has not given you. Say those out loud before you multiply. Where the missing input is a unit of count rather than a ratio, the failure mode is different and sharper, and guesstimate questions cover it.

The market sizing drill returns the four dimensions separately after each attempt, and names the one to fix first, so the pattern you repeat is visible across sessions rather than inferred from a single mark. It is open without an account, as is every figure above.