Market sizing questions, how many piano tuners are in a city, how big is the market for a new product, are a consulting-interview staple precisely because there's no dataset to recall, only a structure to build live. This deck breaks down the toolkit: top-down versus bottom-up thinking, clean population segmentation, sanity-checking your own numbers, and the habits that separate a defensible estimate from a guess. Work through it and the next guesstimate question won't feel like a trick, it'll feel like a format you already know.
17 cardsConsultingFree, no signup
Card 1 of 17
When an interviewer asks you to estimate something like the number of piano tuners in a city, what are they actually evaluating?
They're grading your structure and logic under uncertainty, not whether you nail the exact number.
Market sizing questions test whether you can build a clear, logical structure to estimate an unknown quantity when you don't have the data memorized. Interviewers care about how you break the problem down, state your assumptions out loud, and adjust when something seems off, not whether your final number matches a figure sitting in a database somewhere. A well-reasoned estimate that lands in a plausible range beats a lucky guess with no visible logic every time. The real skill being tested is comfort with ambiguity, can you make steady progress on a problem that hands you no clean, given inputs.
Card 2 of 17
What's the difference between a top-down and a bottom-up approach to market sizing, and when should you use each?
Top-down starts from a big number and narrows it down; bottom-up builds the estimate up from a single unit.
A top-down approach starts with a large, known or estimable figure, like a country's population, and narrows it step by step by applying percentages until you reach the answer. A bottom-up approach does the opposite, it starts from a small, concrete unit, like one store's average daily customers, and multiplies up to the whole market. Top-down tends to work well when you have a solid anchor number to start from, while bottom-up is often more defensible when the top-line figure is fuzzy but a single unit of the business is easy to picture. Strong answers sometimes use both directions on the same question, as a way of cross-checking that the two paths land in the same neighborhood.
Card 3 of 17
Your interviewer says there’s no data and no calculator, then asks you to size a market. Which handful of numbers is worth carrying into every interview, and how do you use them honestly?
Carry a few rough anchors like population, household size and life expectancy, and label them as estimates.
A small set of figures does most of the work: world population is roughly 8 billion, a large developed country runs somewhere around 300 to 350 million people, an average household in a wealthy country holds about 2.5 people, and life expectancy sits near 80 years. Carry them as approximations rather than facts, say out loud that you’re approximating, and round them to whatever makes the arithmetic clean. Nobody expects memorized statistics, but the candidate who reaches for a sensible anchor instead of freezing gets moving in seconds rather than minutes.
Card 4 of 17
Ferrow Retail's interviewer asks you to estimate how many people in Corvale, a fictional city of 900,000, buy a birthday cake in a given year. How do you segment the population without double-counting or leaving gaps?
Split the population into buckets that are mutually exclusive and collectively exhaustive, no overlaps, no gaps.
The core discipline is MECE: every person in Corvale should fall into exactly one segment, and every segment together should cover the whole population. For a birthday-cake question, splitting by age band, say under 10, 10 to 64, and 65-plus, works cleanly because everyone has exactly one age; splitting by categories like 'families' and 'young professionals' instead risks double-counting a young professional who is also a parent. Once the segments are clean, you can apply a different purchase-rate assumption to each, for instance a different bakery-purchase rate for a young child's party versus an adult birthday, and add the pieces back up knowing nothing was counted twice or missed entirely.
Card 5 of 17
An interviewer wants the number of people aged 30 to 44 in Corvale, a fictional city of 900,000, and gives you nothing else. How do you get an age band out of a single population number?
Spread the population evenly across a lifespan, then shave the top bands because fewer people reach old age.
Start by assuming people are spread roughly evenly across a normal lifespan. If life expectancy is around 80 years, each single year of age holds about one eightieth of the population, so Corvale’s 900,000 people give roughly 11,000 per year of age and a 15-year band comes to about 170,000. Then correct for reality, since the top of the distribution tapers as fewer people reach old age, so trim the bands above 65 and nudge the younger ones up. Stating the flat assumption first and adjusting it afterwards is faster and far more transparent than pretending you know the real age curve.
Card 6 of 17
You’ve broken a market into eleven customer segments and the interviewer is starting to look impatient. How do you know when segmenting has stopped adding accuracy?
Split a population only where the buying rate genuinely differs, since extra buckets add work, not accuracy.
Segmentation earns its keep when two groups would get meaningfully different assumptions, the way under-tens and adults buy birthday cakes at genuinely different rates. If you’d apply the same rate to two buckets, merging them costs you nothing but saves you time. Two to four segments is usually plenty in an interview, and a long list of thin slices tends to manufacture false precision, a number that looks carefully built but now rests on ten guesses instead of three. The interviewer is judging what you chose to care about, not counting your boxes.
Card 7 of 17
As you work through a multi-step estimate, how should you handle rounding along the way?
Round to clean numbers on purpose, say it out loud, and track which direction each round pushes your answer.
Pick simple, round inputs at each step, for example treating a population of 4.3 million as 4 million, and state clearly that you're doing it. This keeps your mental math fast and keeps the interviewer able to follow your logic in real time instead of getting lost in decimals. It also helps to notice whether you're consistently rounding up or down across several steps; if most of your rounds went down to make the math easy, your final number is probably a slight underestimate, and flagging that directly reads as more rigorous than presenting a rounded chain of assumptions as if it were precise.
Card 8 of 17
A candidate sizing the annual market for replacement bicycle tires counts every adult in the country and stops there. Which two multipliers did they forget, and how much do they matter?
Not everyone in the pool buys, and buyers buy more than once, so multiply by penetration and frequency.
Penetration rate is the share of your starting pool who actually own or buy the thing, and frequency is how many times each of those people buys it in a year. Leaving either one out quietly sets it to 100% or to once, which is how estimates end up off by a factor of five. For tires you might start with 40 million adults, assume 20% ride often enough to wear tires out, giving 8 million riders, then assume each replaces about two a year, landing near 16 million tires. Say both multipliers out loud, since they’re usually the two an interviewer pushes on.
Card 9 of 17
Walk through how you'd structure an estimate for the number of independent coffee shops in Dunmoor, a fictional mid-size city of 500,000 people.
Anchor on population, narrow to regular buyers, then divide by what one shop can realistically serve.
Start from Dunmoor's population of 500,000 and estimate the share who buy coffee from a shop a few times a week or more, say 30%, giving roughly 150,000 regular customers. Next estimate how often each of them visits per week, maybe 3 times, and what fraction of those visits go to an independent shop rather than a chain, say half, which works out to about 225,000 independent-shop visits per week. Finally, estimate how many customers one independent shop can realistically serve in a week, perhaps 1,500, and divide, 225,000 divided by 1,500 lands you at roughly 150 independent coffee shops in Dunmoor. Each step used one clean, stated assumption, which is exactly what makes the answer defensible even though none of the inputs were 'known' facts.
Card 10 of 17
Partway through an estimate you land on a number, say 40,000 hospital beds in Aldmere, a fictional mid-size country. How do you sanity-check that against something you already know?
Convert your estimate to a per-capita rate and compare it to a number you already trust.
Raw totals are hard to judge, but rates are easy to sanity check. Take your estimate, divide it by population to get something like 'beds per 1,000 people,' and compare that to a rough figure you're confident is roughly right, such as knowing a typical developed country sits somewhere around three to five hospital beds per 1,000 people. Hold that loosely, because the spread is wide: roughly 2.5 in the US and UK, nearer 8 in Germany, above 12 in Japan and Korea. The check is for order-of-magnitude errors, not small ones. If your derived rate is wildly outside that plausible range, some step upstream is probably off, and it's far easier to catch that with a familiar per-capita ratio than by staring at a seven-digit total. This kind of check is also worth narrating out loud, since it shows the interviewer you're pressure-testing your own math rather than trusting it blindly.
Card 11 of 17
Northbay Logistics asks how big the market for delivery vans is. Do they mean how many are on the road or how many get bought each year, and how do you get from one to the other?
The installed base is a stock, annual sales are a flow, and the replacement cycle converts between them.
A stock is how many exist right now, the installed base, while a flow is how many are bought or replaced over a period. The two can differ by a factor of ten, so ask which one the question means before you calculate anything. The bridge between them is how long the product lasts: if there are roughly 200,000 delivery vans on the road and a van lasts about ten years, then around 20,000 need replacing each year, plus whatever the fleet is growing by. A parts supplier cares about that annual flow, while a fleet software vendor cares about the stock.
Card 12 of 17
Your interviewer says a market size should come back in dollars, but everything you’ve built so far is counted in users. How do you convert without inventing a price out of thin air?
Chain users to transactions to dollars one multiplier at a time, and name the unit at every step.
Treat it as a chain and keep the unit written beside every number so you can see what each multiplication actually does. Users times visits per user per year gives transactions per year, and transactions times the average spend per transaction gives revenue per year. For a coffee app that might be 150,000 users, 40 visits a year each for 6 million transactions, at 8 dollars a visit, which is about 48 million dollars. Price is the input people invent most carelessly, so anchor it on what you’d genuinely expect to pay and tell the interviewer that’s what you did.
Card 13 of 17
An interviewer asks you to estimate the number of accordion repair technicians in Velmara, a small fictional country you know almost nothing about. What do you do when you have no anchor data at all?
When you lack a direct anchor, build one from a related number you're more confident estimating.
Start with whatever you're most confident about, usually total population, then work through a chain of plausible ratios to get closer to the target: how many households might own an accordion, how often an owned instrument needs professional repair, and how many repairs one technician can realistically handle in a year. Say each assumption out loud and flag which ones you're least sure of, since the interviewer usually cares more about watching you reason through genuine uncertainty than about a lucky guess. It's also completely fair to state a reasonable working assumption about Velmara, like an approximate population, if the interviewer hasn't given you one, that's a normal, expected part of the exercise, not a sign that you're stuck.
Card 14 of 17
A candidate estimates that Bramwell Fitness's target city has 1.2 million gym memberships in a population of 400,000. What went wrong, and what other mistakes commonly wreck an estimate like this?
Watch for nesting errors, unstated assumptions, and any input that's implausible on its face.
That example fails a basic plausibility check, it implies each person holds three memberships, which points to a nesting error, some rate probably got applied twice, or applied to the wrong base, like visits per week mistaken for total members. Other common breakdowns include quietly changing an assumption partway through without saying so, stacking several unstated guesses on top of each other until the interviewer can no longer follow the logic, and feeding in an input that's implausible on its own, like assuming 90% of adults belong to a gym. The fix for all of them is the same habit, say every assumption out loud as you make it, and pause to ask whether the resulting number still sounds like something that could be true in the real world.
Card 15 of 17
You give your number and the interviewer asks which of your assumptions you’d most want to check before showing it to a client. How do you answer that well?
Name the assumption whose plausible range moves the answer most, not the one you happen to like least.
Because the estimate is a chain of multiplications, an input that’s off by a factor of two moves the final answer by a factor of two no matter where in the chain it sits. So rank your assumptions by how wide their plausible range is rather than by how uneasy they made you feel. Penetration rates are usually the worst offender, since 10% and 40% are both arguable and that alone swings the answer four-fold, whereas average household size only runs from about 2.2 to 2.8. Naming that one input, and which way it would move things, is the conversation a real team has around a client model.
Card 16 of 17
You've landed on a final number, say 60,000 gym memberships in Aldmere. What's the last thing you should do before giving that answer to the interviewer?
Zoom out and ask whether the number would surprise a reasonably informed adult, then explain why or adjust.
Before saying the number out loud, translate it into something intuitive, like memberships per adult or per household, and ask yourself whether that rate feels plausible given everyday experience. It also helps to briefly restate the chain of assumptions that produced it, since walking back through the logic often surfaces one step that quietly went too high or too low. If the number still feels off, it's better to flag that honestly, something like 'this feels a bit high, and I think it's because of X,' than to present a suspicious number with total confidence, interviewers consistently rate that kind of self-awareness higher than a falsely confident wrong answer.
Card 17 of 17
Calder Foods is weighing entry into meal kits in Aldmere, a fictional country of 30 million. Size the annual revenue of that market out loud, then tell the interviewer whether you believe it.
Chain households, penetration, frequency and price, then check the total as a per-person number.
Thirty million people at roughly 2.5 per household gives 12 million households. Say 30% are urban and affluent enough to be plausible buyers, and that 10% of those actually subscribe, leaving about 360,000 subscribing households. Each takes two kits a week for 50 weeks, so roughly 36 million kits a year, and at 30 dollars a kit the market is a little over a billion dollars. Then check it before you stop: that’s about 35 dollars per person per year against maybe 3,000 dollars of annual grocery spend, close to 1% of the food budget, which is plausible for a category that’s still a niche.
Reading is not the same as remembering. Study this deck with scheduled review so the cards come back before you forget them.