Case interviews: structure, frameworks and a worked sample

What a case interview measures

MIT’s career office notes that case interviews are a common part of the process when interviewing for consulting positions. The format has since spread to strategy, operations and product roles in other industries.

The thing being measured is not business trivia. Harvard’s guidance frames case interviews as a test of analytical and problem-solving ability rather than deep business knowledge, and states the standard directly: it is not the solutions they are looking for, it is the analysis. A candidate who reaches a wrong recommendation through a clear, well-sequenced argument usually outscores one who guesses the industry-standard answer and cannot explain how they got there.

Four things are being watched: whether you structure a vague problem into parts that do not overlap, whether you ask for information instead of assuming it, whether your arithmetic holds up out loud, and whether you can state a recommendation with its main risk attached.

It is also a test of composure, since you have to think with an audience. Practice is what makes that bearable, which is why this page ends with how to practise rather than more theory.

The first three minutes

Harvard’s guidance gives an explicit opening sequence: repeat the question, ask a few clarifying questions, organise your thoughts. Almost every case that goes badly goes badly because one of those three was skipped.

Repeating the question takes fifteen seconds and does two jobs: it confirms you heard the actual prompt, and it buys you a moment to think. Say it back in your own words, including the numbers.

Clarifying questions come next, and three is usually the right number. Ask what the client means by the goal, since “grow” can mean revenue, profit or share. Ask over what time frame. Ask what they have already tried, because that often rules out half the obvious answers. Twelve questions reads as fishing.

Then structure, out loud, before you analyse anything. Name the two to four parts you intend to examine, say why they cover the problem, and ask whether that sounds reasonable. This is the step candidates skip under pressure, and the step the interviewer is most clearly scoring, because the structure is the analysis. Asking for thirty seconds to lay out your approach is normal, and far better than narrating half-formed thoughts.

Frameworks, and why a borrowed one fails

A framework is just a way of cutting a problem into parts that do not overlap, and a handful of standard cuts cover most cases. Profit splits into revenue and costs, revenue into price and volume, costs into fixed and variable. Market questions split into customers, competitors, company and product. An entry decision splits into whether the market is attractive, whether this company can win in it, and whether the economics work at scale.

What gets candidates rejected is reciting one of these and then fitting the case to it. Asked why a hospital’s patient wait times have risen, a candidate who responds with a four-part market framework has shown only that they memorised something. Build the cut from the question in front of you, using the standard cuts as raw material.

A good structure has three properties: the parts do not overlap, together they plausibly cover the problem, and each part can be investigated with information you could ask for. A branch with no way to be tested is decoration.

A worked case study interview sample

The case below is invented for practice, and the figures are illustrative rather than drawn from any real company. That is also true of the cases you will be given: the interviewer supplies the numbers as you ask for them.

Interviewer: Our client runs a regional chain of 40 bakeries. Over the last two years, operating profit has fallen by about 20 percent while revenue has stayed flat. They want to know why. Where would you start?

Candidate: So the client has 40 bakeries, revenue is flat across two years, and operating profit is down roughly 20 percent. The question is what is driving the profit decline. Before I structure it, three quick questions. Is the store count the same across both years? Has the product mix changed? And is the decline spread across stores or concentrated in some of them?

Interviewer: Store count is unchanged, mix is broadly the same, and the decline is spread fairly evenly.

Candidate: Then since profit is revenue minus costs, and revenue is flat, the cause has to be on the cost side. I would like to split costs into the main buckets, find which one grew, and then work out whether the growth is in price per unit or in quantity used. Does that sound reasonable?

Interviewer: It does. Revenue is 50 million dollars in both years. Costs were 45 million two years ago and 46 million now.

Candidate: That is consistent: profit of 5 million falling to 4 million is the 20 percent decline. So I am looking for a 1 million dollar cost increase. Could I have the split of those costs?

Interviewer: Ingredients were 20 million and are still 20 million. Rent is 6 million in both years. Other costs are 4 million in both years. Labour was 15 million and is now 16 million.

Candidate: Then the entire increase is labour. With store count flat and ingredient spend flat, volume has probably not moved much, so I would expect either more hours worked or a higher cost per hour. Do we know which?

Interviewer: Total hours are roughly unchanged.

Candidate: So this is a wage rate effect: about a 7 percent rise in cost per hour with no change in hours or output. My recommendation would therefore focus on labour productivity and pricing rather than on sales. The client has absorbed a wage increase without passing any of it through, and with ingredients and rent flat there is no cost offset available elsewhere. The main thing I would want to test before committing is price sensitivity, because a 2 percent price rise on 50 million dollars of revenue would recover the full million, and whether that is achievable depends on local competition.

Notice what the candidate never does: guess. Every number comes from a question, every step is stated before it is taken, and the answer names the risk that would break it.

Market sizing and doing arithmetic out loud

Many cases contain an estimation question. The method is always the same: build the number from a chain of assumptions you state as you go, keep them roundable, and sanity-check the answer at the end.

Suppose you are asked to estimate the annual revenue of a single city-centre coffee shop. Start with time: say the shop opens twelve hours a day. Split those hours, because the traffic is not even: three peak hours at roughly 100 customers an hour, and nine quieter hours at roughly 25 an hour. That is 300 plus 225, so call it 500 customers a day. At an average spend of 6 dollars, the shop takes about 3,000 dollars a day. Over 350 trading days, that is around 1.05 million dollars a year, which you would round to about one million.

Harvard’s guidance includes a practical tip for exactly this moment: round numbers so the mental arithmetic stays manageable. Carrying 487 customers and 5.83 dollars through four multiplications is how candidates lose the thread. Round early, say that you are rounding, and keep track of the direction you rounded in.

Then challenge your own result, out loud. One million dollars a year for a busy coffee shop is the right order of magnitude; had your assumptions produced 50 million, a step was wrong. Noticing an implausible answer is itself part of what is being scored.

These figures are illustrative assumptions, not measured data. In a real case the interviewer either gives you the inputs or accepts yours, provided you justify them.

The four case formats you may meet

A candidate-led case is the classic version: a short prompt, and you drive, asking for data and choosing the next step. Structure and initiative carry the most weight.

An interviewer-led case moves in defined stages, where the interviewer asks a question, takes your answer, then hands you the next one. You do less steering, but each answer is scored more tightly and the arithmetic is more exact.

A written case gives you a pack of exhibits and some preparation time, then asks you to present and defend a recommendation. The skill shifts toward noticing which chart actually answers the question.

A group case puts several candidates on one problem, so the assessment becomes partly behavioural: whether you build on other people’s analysis or talk over it. If that is your format, read the group interview guidance too. Either way, ask the recruiter which format you are getting; most will tell you.

How to practise, and the mistakes to avoid

Cases cannot be learned by reading. Work through them out loud with a partner playing the interviewer, because the gap between thinking clearly and speaking clearly is the whole difficulty.

Build the arithmetic separately. Practise multiplying round numbers in your head and writing a calculation on paper while still talking.

Four mistakes account for most failures. Analysing before stating a structure, which leaves the interviewer unable to follow you. Assuming facts instead of asking for them, the error the format is designed to expose. Going silent, since nobody can score reasoning they cannot hear. And stopping at the analysis with no recommendation.

One last habit: when you notice you have gone down a dead end, say so and redirect. Admitting that a branch is not explaining the problem is read as judgement. Quietly continuing down it is read as the opposite.