Guesstimate Interview Questions: Count the Right Unit

One firm names guesstimate interview questions, and what it watches

Most firms describe case interviews without ever using the word. Charles River Associates uses it, in its own spelling, on its interview tips page: case interviews "are designed to test how you approach a problem. There's usually no single correct answer ("guestimations" are common), but what we are looking for is how you react to situations with limited information; how you apply logic and reasoning to the question; what data you draw on; and how you communicate to bring the interviewer along with your thinking."

The same paragraph ends with an instruction that sits oddly next to "no single correct answer" until you read it twice: "be prepared to explain how you came to your answer and show your math."

Both halves are true at once. The figure is not marked. The route to it is, and an unshowable route cannot be marked at all. That is the whole brief for this question type, written by a firm, and it explains why the two failures below cost so much. Neither is an arithmetic slip. Both corrupt the route.

Four guesstimate interview question types, and the unit each counts

Four families cover almost everything asked under this label, and the useful way to sort them is by the unit you have to declare before the first multiplication. The sorting is ours. Charles River Associates is the one firm in our reading that names the format, and it describes it without dividing it up.

Row one is the wedding error in the next section, where two participants make one occasion. Row two is the packing factor below, where a sphere occupies about 52 percent of the cube you drew around it. Row three breaks when a per-hour rate gets multiplied by twenty-four instead of by the hours a business actually runs, which on a shop open ten hours a day overstates the answer by a factor of 2.4. Row four is visible in the Arthur D. Little exercise further down: 360 EUR a year buys one subscription, and a gym membership shared between two adults is one account with two users.

Name the family out loud and the unit sentence writes itself. "This is a throughput question, so I am counting cars through one lane in one opening hour." Thirty seconds of sorting removes the error class the rest of this page is about.

The unit error, told by the person who made it

A first-year associate in a firm's London office wrote up his own interview on his employer's careers blog, and the most useful line in it is a mistake, the kind that also appears on the list a firm publishes about its own candidates. Sizing a wedding market, he counted each participant as a separate event. The interviewer's reply is quoted directly in his account: "Well, that's great methodology, but are you forgetting something?"

Put numbers on it. Suppose 1.2 million people get married in a year. That is 600,000 weddings, because a wedding takes two people. Size the market at £25,000 per wedding and you get 600,000 times £25,000, which is £15bn. Size it per person and you get 1.2 million times £25,000, which is £30bn. The chain was identical. The answer is out by a factor of two.

His conclusion is worth keeping too. "The key was not necessarily getting the right answer but demonstrating structured problem-solving skills." He passed. He is a named first-year associate at the firm, which is what makes the anecdote citable rather than folklore. One account is an account and not a pattern, and we treat it as his experience rather than as firm policy.

The general form of the error is worth stating because it recurs in every direction. Per household against per person. Per event against per participant. Per vehicle against per passenger. Per subscription against per user. Each pair differs by a factor between two and four, and the factor never announces itself in the arithmetic. Say your unit out loud before the first multiplication, in the form "I am counting weddings, at two people each." That sentence costs four seconds and it is the single cheapest error check in this question type.

A volume guesstimate where the packing factor decides

Volume guesstimates are the family where geometry decides the answer, so here is one all the way through. How many golf balls fit inside a twenty-foot shipping container?

Two declared assumptions, because nothing here is given. Call the inside of the container 5.9 m long, 2.35 m wide and 2.39 m high. That is 5.9 times 2.35, which is 13.87, times 2.39, so about 33 cubic meters. And call a golf ball 4.3 cm across.

The fast route. Give each ball a 4.3 cm cube to itself. A cube that size holds 4.3 cubed, which is 79.5 cubic centimeters. The container holds 33 million cubic centimeters, and 33,000,000 divided by 79.5 is 415,000. So roughly 400,000 balls, and you can say that in twenty seconds.

The honest route. The cube method has already made a packing assumption without telling you. A sphere fills π over 6 of its bounding cube, which is about 52 percent, so stacking balls in a cubic lattice wastes nearly half the space. Poured in loose, equal spheres settle at about 64 percent, which is better than a cubic lattice. A ball's own volume is π over 6 times 4.3 cubed, so 41.6 cubic centimeters. Take 33 cubic meters times 0.64, which is 21.1 cubic meters, or 21,100,000 cubic centimeters, and divide by 41.6. That gives roughly 508,000.

The wrong route, and it is the common one. Divide container volume by ball volume with no packing factor at all: 33,000,000 over 41.6 is 793,000. That is 56 percent above the 64 percent figure, because one over 0.64 is 1.5625.

The first two answers differ by about 22 percent, which is well inside the noise of a guesstimate. The third is a different kind of wrong, because it asserts something physically impossible. The scoring difference between the first two rows is zero. The difference between either of them and the third row is the whole question.

State the packing factor whichever route you take. "Call it 400,000, and that already assumes each ball gets a cube, which is conservative" is a complete answer. The number on its own is not.

When the guesstimate interview question does not close

Arthur D. Little publishes a quantitative exercise on its case study page, and the numbers in it cannot all be true. The text reads: a city of "5 million citizens and a market size of 300 million EUR per year", where the client "hopes to cover 25% of the budget market and requires at least 210 million EUR of revenue per year to do so", with "budget gyms' clients spending an average of 30 EUR per month (360 EUR yearly)".

Twenty-five percent of 300 million is 75 million. The figure given is 210 million, which is 70 percent of 300 million. Both readings are computable and they do not agree.

The two subscriber counts differ by a factor of 2.8, which is exactly 210 over 75. And the question the page actually asks cannot be answered on either reading: it asks this. "What percentage of people currently not working out need to join Penny Pressure to reach the target revenue level?" The current participation rate appears nowhere on the page. Without it the denominator does not exist. The same careers site asks for a structure with at least two layers, and for time to think before you present it, which is where the missing input gets raised.

We do not decide which figure the firm meant. What the situation teaches is the response, and the firm's own page supplies it two lines below the prompt, advising candidates not to be afraid to ask clarifying questions. Say the contradiction out loud, pick one reading, name it, and carry on. "Twenty-five percent of 300 million is 75 million, and you have given me 210, so I will work with 210 and flag that it implies 70 percent of the market." That answer beats a silent choice even when the silent choice is the right one.

Five moves, in the order that survives a wrong turn

Charles River Associates publishes the brief for this question type in one sentence. It watches how you react to situations with limited information, how you apply logic and reasoning, what data you draw on, and how you communicate to bring the interviewer along. Four things, and only the third of them is a number. The order below is ours, and each move exists to protect one of the four dimensions our grader scores.

1. Name the family and the unit. Count weddings, at two people each. Protects Scope, worth 25.
2. Ask the one question that changes the chain. The Arthur D. Little page advises candidates not to be afraid to ask clarifying questions, and the contradictory gym figures below are what the silence costs. Protects Scope again, while it is still free to fix.
3. Write the chain before computing it. Population, then share, then frequency, then price, each on its own line with its unit beside it. Protects Assumption chain, worth 30, because a chain you can see is a chain where a challenged input costs one line to repair.
4. Compute out loud, one operation per sentence. The same firm page asks you to show your math. Protects Arithmetic, worth 25, which is scored on whether the multiplication was visible.
5. Close on a physical bound. Weight, area, seats, hours in a day. Protects Sanity check, worth 20, the dimension that takes no partial credit.

The order earns its place in one specific way. Moves one and two are the only two that are still free. After move three the chain exists on paper, and changing the unit then means doing every operation twice. On a four-step chain that is four multiplications repeated, which at three minutes a calculation is most of the time a guesstimate gets.

The four notation habits that prevent lost zeros

Arthur D. Little prints one of these itself, advising candidates to write 1M in place of the long form 1,000,000 when a calculation needs many operations in a row. The other three are ours, and together they remove the most common cause of a wildly wrong guesstimate.

1. Work in one scale and name it once. Millions, or thousands, chosen at the top and written at the top of the page.
2. Write the unit beside every running total. Write "kL/day" or "balls" in the margin. A quantity with no unit written next to it is the one that gets multiplied by the wrong thing.
3. Convert at the boundary, never mid-chain. Centimeters to meters, bottles to gallons, months to years. Do the conversion as its own line with its own result.
4. Restore the scale last, and check the order of magnitude against something physical. Half a million golf balls at 45 g each is 22,500 kg, so roughly 22 tonnes. That is a heavy but plausible container load, which means the answer survives a physical check. An answer of five million balls would have implied 225 tonnes and failed it.

That fourth habit is the sanity check this question type rewards most, because a guesstimate has no market figure to test against. A physical bound will usually do instead: weight, area, hours in a day, seats in a room.

How we score a guesstimate, and whose scale this is

Guesstimates run on the same four dimensions as any estimate on this site. The weights come from src/lib/grading/rubrics.ts, which means they are ours. No firm publishes a marking scheme for an estimate, and neither of the two firms with published estimation exercises above provides an answer key of any kind.

Five anchors, and the bounds nobody prepares

Five is enough, and they have to be for your market. A national population, a household count, the number of households per dwelling type, working days in a year, and one price you actually know. Look them up once at a national statistics office so that they are figures rather than impressions, then stop collecting. A sixth anchor adds less than one more practice question.

What does keep paying off is the physical set, because it never goes out of date. The volume of a sphere as π over 6 of its bounding cube. Random close packing at about 64 percent. A cubic meter as a thousand liters. A liter of water as a kilogram. Four of those cover most volume guesstimates you will meet.

There is one more class of anchor worth the effort, and it is the one nobody prepares: the bound. Seats in a cinema, cars through a lane in an hour, people through a turnstile in a minute, hours a shop is open. Bounds are useful because they cap an answer from above without any chain at all. If the guesstimate you built implies forty thousand people through one entrance in a day, a bound tells you immediately that the entrance count is wrong, and it tells you before you have committed to a figure. Collect three or four bounds for the kind of business your target firm serves and you will catch the errors a sanity check would otherwise find too late.

Six prompts to run the five moves on

Six prompts, drawn from the first three families, and none of them needs a figure you cannot look up once at a national statistics office. They are ours. Work them in the order below and declare the unit before touching a number.

Two of the six worked through, so the set is not a list of questions with nothing behind it. On the turnstile, a peak of two hours at one entry every two seconds gives 1,800 entries an hour, because 3,600 seconds divided by 2 is 1,800. Two hours through one gate is 3,600 entries. On the brick wall, a face 20 m long and 12 m high is 240 square meters, and a standard brick face of 215 mm by 65 mm is 0.013975 square meters, which we will round to 0.014 and say so. Then 240 divided by 0.014 is about 17,100 bricks for that face.

Recomputed by hand on 7 October 2026. Both figures are the first line of an answer. Neither is finished until the bound is stated. A face of 17,100 bricks spread over four stories is 4,275 a story, which rounds to roughly 4,300, and at an assumed 500 bricks a day that is 8.6 days of one bricklayer's work. A plausible number is what a bound is for.

Drilling the unit sentence

Do the unit sentence on every prompt before anything else, even when the unit is obvious, because the whole point is that it stops being a decision. Then do the closing sanity check, which is the dimension with no partial credit.

The market sizing drill runs the four dimensions above and reports each separately, so a scope miss stops looking like an arithmetic miss. Where the question has a real market behind it and the failure moves to scope and denominator, market sizing questions covers the differences. Where the estimate feeds an investment decision, break-even analysis picks up from the volume you produced.

Then practice the part that feels wasteful, which is saying the assumption out loud every time. A packing factor you know and do not mention scores the same as a packing factor you never considered, because the grader and the interviewer both see the same thing: a number that arrived without a justification. Recording yourself for one session is the fastest way to find out how much of your reasoning stays silent. Most people are surprised by how much.

One thing this page cannot give you, and it is worth naming rather than hiding. Whether a particular firm asks guesstimates at all is a question only that firm answers, and most of them do not publish it. Charles River Associates says they are common in its process. The other firms whose preparation pages we read describe cases without mentioning the format either way. Ask your recruiter when the invitation arrives, and if you get no answer, prepare for it anyway: the unit sentence and the sanity check pay off in every other case type too.

The weights above and the reasoning behind them sit on how an answer is assessed. Everything on this page is reproducible from the three firm pages linked in the text, and none of it sits behind a sign-up.