Case interview examples: five prompts, and how to work them

Where can I find case interview examples, and how should I use them?

Case interview examples are practice prompts you work under timed conditions, not solutions you read. Use one properly: sixty seconds clarifying, ninety seconds structuring out loud, then the analysis, then mark yourself against the model. This guide gives five prompts we wrote in-house, one each on profitability, market entry, pricing, market sizing and capacity, with the full working for every number.

What to take away

  • Reading a worked solution builds recognition, not fluency: the value of an example is destroyed the moment you skip ahead to the answer
  • The protocol is 60 seconds clarifying, 90 seconds structuring out loud, then the analysis, then a self-mark against the model
  • In a profitability case, isolate revenue versus cost before proposing anything: prescribing before diagnosing is the single most common failure
  • A market-entry answer that sizes the prize but never tests the right to win is half an answer, and interviewers mark it as half
  • The breakeven volume loss on a price rise is the price change divided by the new contribution margin, so a 10% rise on a 70% margin tolerates 12.5%, not 14.3%
  • In a capacity case, quantify the gap in units per day before ranking any option: the size of the gap decides which options are even relevant

How to use an example case without wasting it

Most candidates consume example cases the way they consume recipe videos. They read the prompt, feel a flicker of an idea, then scroll down to the model answer, nod, and file the case away as "done". Three weeks later they cannot reproduce any of it under time pressure, and they conclude they need more examples. They do not. They need to use the ones they have differently.

The reason is worth being precise about. Reading a solution trains recognition: you become able to recognise a good structure when you see one. A case interview tests production: whether you can generate a structure, out loud, in ninety seconds, for a problem you have not seen. Those are different skills and one does not convert into the other. This is why the candidate who has read forty case solutions often performs worse than the candidate who has properly worked five.

So there is one rule that governs everything below. Work the prompt before you read a single line of the working. If you break that rule, the example is spent: you cannot un-see the answer, and you will never know whether you would have got there. Treat each of the five prompts in this guide as a one-use resource.

The protocol

01

Sixty seconds of clarifying

Read the prompt once, then ask, out loud, the questions you would ask a real interviewer. What is the client's objective, and by when? What are the units and the currency? Is there a constraint I should know about? Do not ask more than three, and never ask a question whose answer you could reason to yourself. Then say the objective back in one sentence.

02

Ninety seconds of structuring

Silence is allowed and expected. Build two to four branches that fit this problem, name what sits under each, then say which branch you want to start with and why. The last part is the part candidates skip, and it is the part that carries the mark. A structure with no stated starting point reads as a list rather than a plan.

03

Work it out loud, with a timer

Fifteen to twenty minutes on the clock. Every calculation spoken as you do it: the assumption, the arithmetic, the answer, then a sense-check. If you find yourself thinking in silence for more than about twenty seconds, say what you are weighing. An interviewer cannot mark reasoning they cannot hear.

04

Close with a recommendation in sixty seconds

Answer first, then two or three reasons with numbers attached, then the biggest risk, then the next step. Do this before you look at anything below. A recommendation you compose after reading the model answer is not evidence of anything.

05

Then compare, and mark yourself

Now read the working. Score yourself 1 to 5 on framing, rigor and synthesis, writing one line of evidence from your own recording before you write the number. If you cannot point to the moment, the score does not stand.

One honest note on what follows. The five prompts below are written by us, for practice. They are not transcripts of real client work and they are not McKinsey, BCG or Bain cases, because those firms do not publish theirs. The companies are invented. What is real is the shape of the problem, the arithmetic, and the standard the working is held to.

Example one · profitability: why has a bakery chain's profit fallen?

Profitability is the most common case shape you will meet, and the one candidates most often fumble by moving too fast. Work this prompt in full before reading on.

The prompt

Thornbury Bakehouse is a regional chain of 48 bakery cafes in the north of England. Two years ago it earned £6M of operating profit. Last year it earned £3.6M. Revenue has not moved.

The managing director wants to know what has happened and what she should do about it. Where do you want to start?

The model structure

Profit is revenue minus cost, so the first branch of the tree is not a framework choice, it is arithmetic. Revenue is flat and profit has fallen, therefore cost has risen by exactly the amount profit fell. You can say that in the first fifteen seconds and it immediately halves the problem. From there the structure has three levels: is the cost rise in cost of goods, labour, property or overhead; is it price per input or quantity of input; and is it a chain-wide effect or a few bad sites.

The move that separates a strong opening from a weak one is asking for the numbers in per-unit terms rather than totals. Totals tell you where the money went. Per-unit figures tell you whether the business got more expensive to run or simply ran more.

Working the diagnosis
Profit fell £6.0M to £3.6M, so the fall is £2.4M. Margin fell from 6.0 ÷ 60 = 10% to 3.6 ÷ 60 = 6%.
Revenue is flat at £60M, so total cost rose from £54M to £56.4M. That is the whole £2.4M, and it is a 4.4% cost increase on a flat top line.
Volume: 30M items sold at an average £2.00. Price is unchanged and revenue is unchanged, so volume is unchanged. Nothing here is a demand problem.
Cost per item therefore rose from 54 ÷ 30 = £1.80 to 56.4 ÷ 30 = £1.88, an increase of 8p per item.
Ask for the per-item cost lines. Two years ago: ingredients £0.60, in-store labour £0.72, property £0.30, other overhead £0.18. Sum: £1.80. Checks out.
Last year: ingredients £0.66, labour £0.74, property £0.30, overhead £0.18. Sum: £1.88. Checks out.
Ingredients: +£0.06 × 30M items = £1.8M. Labour: +£0.02 × 30M = £0.6M. Property and overhead: unchanged.
£1.8M + £0.6M = £2.4M, which is the entire profit fall. Ingredients alone are 1.8 ÷ 2.4 = 75% of it.
Sense-check: ingredient cost per item rose 0.06 ÷ 0.60 = 10% in two years. That is high but not absurd for flour, butter and eggs, so the story holds.
Sizing the fix: recovering £2.4M on 30M items needs 8p per item, which on a £2.00 average price is a 4% price rise, if volume holds.

What a 5 does differently

A candidate scoring 3 gets to "it is a cost problem, and mostly ingredients". A candidate scoring 5 does three more things. They notice that flat revenue on flat volume with rising input costs means the chain has absorbed 10% input inflation without passing any of it through, which is a pricing decision rather than a cost accident. They ask whether the 8p is uniform across sites, because a chain-wide input rise and eight badly run cafes need completely different responses. And they attach the 4% price rise to the risk that matters: a bakery cafe competes on a £2 impulse purchase, so a 4% rise may cost volume, and the answer needs a view on how much volume it can afford to lose.

Example two · market entry: should an appliance maker enter a new country?

Market entry rewards discipline over breadth. There are three questions and only three, and candidates lose marks by answering the first one at great length and the other two not at all.

The prompt

Vanderhelm Appliances is a mid-sized European maker of built-in kitchen ovens and hobs. It sells in six countries and is considering a seventh: a northern European market of 10 million people where it has no presence.

The board has approved a business case that assumes 8% market share by year three. They want your view on whether to go. All figures are the ones I give you.

The model structure

Three branches, in this order. Market attractiveness: how big is the prize, how fast is it growing, and how is it structured between premium and value. Right to win: does the client have something the incumbents do not, and can it get to the customer at all, which in built-in appliances means retail and kitchen-installer relationships rather than advertising. Entry economics: what does it cost to be present, what does each unit contribute, and at what share does the venture pay for itself.

State the order out loud and say why, because the order is itself a piece of judgement: the size of the market is the least decision-relevant of the three. A large market you cannot win in is worse than a small one you can, since you will spend real money finding out.

Sizing the market, then testing the plan against it
Population 10M, average household 2.5 people, so 10 ÷ 2.5 = 4M households.
Assume 70% of households have a fitted kitchen with a built-in oven: 4M × 0.70 = 2.8M installed ovens.
Replacement cycle 10 years, so replacement demand = 2.8M ÷ 10 = 280,000 units a year.
New installations: assume 1% of households fit a new kitchen each year = 4M × 0.01 = 40,000 units.
Annual market = 280,000 + 40,000 = 320,000 ovens.
At an average retail price of €450, that is 320,000 × 450 = €144M at retail.
Now test the board's plan. 8% share = 320,000 × 0.08 = 25,600 units.
The client sells to retailers at €270, roughly 60% of retail. Revenue = 25,600 × 270 = €6.9M.
Contribution margin is 30% of wholesale, so €81 a unit. Contribution = 25,600 × 81 = €2.07M.
Fixed cost of being present, per year: sales team, certification, marketing, service network = €3.0M.
So the plan loses about €0.9M a year at its own target share.
Breakeven units = 3,000,000 ÷ 81 = 37,000 units. As a share: 37,000 ÷ 320,000 = 11.6%.
The finding: the board approved 8% share, and the venture needs 11.6% to break even. The plan is not ambitious, it is short.

What a 5 does differently

The sizing above is unremarkable and every prepared candidate can do it. The mark is earned in the last three lines, where the sizing is turned back on the client's own assumption and produces a number the board did not have. That is what commercial judgement looks like on a scorecard: not a bigger market estimate, but the observation that 8% and 11.6% are on opposite sides of a line.

A strong candidate then goes after the right to win with the same specificity. Who holds the shelf space, and are those relationships exclusive. Is the market premium or value, because a mid-sized European maker with no local brand recognition has a plausible story in premium and almost none in value. Is there a partnership or acquisition route that buys distribution rather than building it, which changes the €3.0M fixed line and the breakeven share with it. And they name the one piece of evidence that would settle it, rather than listing everything they would like to know.

Example three · pricing: should a software company raise its subscription price?

Pricing cases look like judgement questions and are really arithmetic questions with a judgement attached at the end. There is one calculation the interviewer is waiting for, and candidates get its denominator wrong more often than any other number in case interviewing.

The prompt

Larkfield Systems sells scheduling software to independent veterinary practices. It has 40,000 subscribers paying £1,200 a year. Contribution margin is 70%.

The CEO wants to raise the price by 10% and is worried about churn. She has asked you: how much churn can we take before this is a bad idea?

The model structure

Say the structure in one breath: I want to work out the breakeven volume loss, then form a view on whether actual churn will land inside it, then check what the price rise does to things the arithmetic does not capture. Then do the maths.

Breakeven volume loss on a 10% price rise
Today: 40,000 × £1,200 = £48M revenue. Contribution at 70% = £33.6M.
Per subscriber: contribution £840, so variable cost is 1,200 − 840 = £360.
After a 10% rise the price is £1,320. Variable cost does not change, so contribution per subscriber becomes 1,320 − 360 = £960.
Breakeven subscribers = £33.6M ÷ £960 = 35,000.
Loss tolerated = 40,000 − 35,000 = 5,000 subscribers, which is 5,000 ÷ 40,000 = 12.5%.
As a formula: tolerable volume loss = price change ÷ new contribution margin = 0.10 ÷ 0.80 = 12.5%.
The denominator is 0.80 because contribution per unit, measured against the OLD price, goes from 0.70 to 0.70 + 0.10 = 0.80. Every extra penny of price is pure contribution.
Using the old 0.70 gives 0.10 ÷ 0.70 = 14.3%, which overstates the tolerance by nearly two percentage points of subscribers. That is the standard error on this case.
Sense-check the upside: if churn only costs 4% of subscribers, 38,400 × £960 = £36.9M contribution, up £3.3M on £33.6M.

Say the denominator out loud when you use it. Interviewers listen for whether you know why it is 0.80, because a candidate who has memorised the formula and a candidate who has understood it are indistinguishable until this moment.

What a 5 does differently

A 3 produces 12.5% and stops. A 5 produces 12.5% and then says the sentence that matters: this is a threshold, not a forecast. Twelve and a half per cent is what the business can survive, and the actual question is what churn will be, which the arithmetic cannot tell you. So they reach for the evidence that could: what happened at the last price rise, how much of the base is on multi-year contracts and therefore cannot leave this year, what a practice pays for the nearest alternative, and how much of the £1,200 is embedded in the customer's daily workflow.

Then they name the asymmetry. A price rise that loses 5% of subscribers is a good trade on contribution and a bad trade on installed base, because the subscribers who leave are cheapest to lose today and most expensive to reacquire later. And they distinguish between raising the price for new customers and raising it for the existing 40,000, which is the same 10% with two completely different churn profiles.

Example four · market sizing: how many professional haircuts does a city buy each year?

Market sizing is the one case type where the answer genuinely does not matter and the method genuinely does. Nobody knows the true figure, the interviewer included. What is being marked is whether your chain of assumptions is visible, defensible and sense-checked.

The prompt

Estimate the annual number of professional haircuts bought in a city of 2 million people.

Take whatever assumptions you need. Talk me through them as you go.

The model structure

Build bottom-up from population, and segment on the variable that actually moves the answer. Here that is haircut frequency, which differs by a factor of two or more between men and women, so a gender split earns its place in the structure. An age split earns its place too, but only one cut: children have a different frequency and a different decision-maker. Do not segment on anything you cannot then put a number against.

Announce the shape before you start calculating: population, then segments, then penetration within each segment, then frequency, then multiply and add. Round every assumption to something you can do in your head. Precision you cannot justify is worse than roundness you can.

Bottom-up chain, with a sense-check at the end
Population 2,000,000. Assume 15% are under 15: 300,000 children, 1,700,000 aged 15 and over.
Split the adults evenly: 850,000 men, 850,000 women.
Men: assume 70% use a barber or salon rather than cutting at home. 850,000 × 0.70 = 595,000.
Frequency for men: every six to seven weeks, so 8 cuts a year. 595,000 × 8 = 4,760,000 cuts.
Women: assume 60% pay for a cut. 850,000 × 0.60 = 510,000.
Frequency for women: 4 cuts a year, longer intervals and often longer appointments. 510,000 × 4 = 2,040,000 cuts.
Children: assume 50% get a professional cut, at 5 a year. 300,000 × 0.50 = 150,000, and 150,000 × 5 = 750,000 cuts.
Total = 4,760,000 + 2,040,000 + 750,000 = 7,550,000 cuts a year. Call it 7.5M.
Demand-side sense-check: 7,550,000 ÷ 2,000,000 = 3.8 cuts per resident per year. For a population that includes non-payers, that sits about where it should.
Supply-side sense-check: a stylist doing 6 cuts a day, 5 days a week, 46 weeks a year does 1,380 cuts a year. 7,550,000 ÷ 1,380 ≈ 5,500 stylists.
That is one working stylist per 2,000,000 ÷ 5,500 ≈ 365 residents. Plausible for a city, and it is the check that would have caught an order-of-magnitude error.
If asked for value: at an average £22 a cut, 7,550,000 × 22 = £166M a year.

What a 5 does differently

The supply-side check is the whole difference. Almost every candidate can build the demand chain. Very few then rebuild the answer from the other direction and see whether the two meet, and that single move is what turns a plausible number into a defended one. It is also the move that saves you when you have slipped a decimal place, which happens to everyone eventually.

A 5 also flags which assumption the answer is most sensitive to before being asked. Here it is male frequency: eight cuts a year against six would move the total by about 1.2M cuts, roughly 16%, while a ten-point change in the children's penetration moves it by 150,000, around 2%. Knowing which of your own assumptions is load-bearing is what lets you defend the estimate under pushback instead of rebuilding it.

Example five · operations: what does a distribution centre do when it runs out of capacity?

Operations cases are underpractised relative to how often they appear, and they punish a specific instinct: reaching for the biggest solution before measuring the problem. Work this one before reading the numbers.

The prompt

Kestrel Household Goods ships through a single distribution centre. It handled 12M cartons last year and the commercial team is forecasting 15M next year on the back of two new retail accounts.

The operations director has put a new distribution centre in the capital plan. The CFO wants a second opinion before it goes to the board.

The model structure

Two branches, and the first one is not optional. Size the gap: what can the site do per day today, what will it need to do, and what is the difference. Close the gap: rank every option by throughput added, capital required and lead time, then check whether the ranking survives the growth continuing past next year. The reason the gap comes first is that its size determines which options are even on the list. A 14% shortfall and a 60% shortfall have almost nothing in common as problems.

Sizing the capacity gap, then ranking the options
The site runs 250 days a year. Current throughput = 12,000,000 ÷ 250 = 48,000 cartons a day.
Required next year = 15,000,000 ÷ 250 = 60,000 cartons a day. Growth is 25%.
Capacity: 30 pick stations at 110 cartons an hour, over two 8-hour shifts. 30 × 110 = 3,300 an hour, × 16 hours = 52,800 cartons a day.
So the site runs at 48,000 ÷ 52,800 = 91% utilisation today. It is tight but not broken.
Gap = 60,000 − 52,800 = 7,200 cartons a day. Against capacity that is 7,200 ÷ 52,800 = 13.6%; against the requirement it is 7,200 ÷ 60,000 = 12%.
That is the headline: the shortfall is 12 to 14 per cent, not a doubling. Hold that number against every option below.
Option A, re-slot the warehouse to lift the rate from 110 to 120 an hour: 30 × 120 × 16 = 57,600 a day. Adds 4,800, which is 4,800 ÷ 7,200 = two thirds of the gap. Weeks of lead time, almost no capital.
Option B, extend the two shifts: at the improved 3,600 an hour, the remaining 2,400 needs 2,400 ÷ 3,600 = 40 minutes of overtime a day. Immediate, pure labour cost.
Option C, add 5 pick stations in existing floor space: 35 × 110 × 16 = 61,600 a day, covering the gap with 1,600 a day spare. Months of lead time, modest capital.
Option D, a new distribution centre: solves everything and more, largest capital, 18 to 24 months of lead time. It cannot deliver next year at all.
A and B together get to 60,000 a day for weeks of work and no capital. That is the answer for next year.
OptionAdded per dayLead time
Re-slot for a higher pick rate+4,800 cartonsWeeks
Extend shifts by 40 minutes+2,400 cartonsImmediate
Add 5 pick stations+8,800 cartonsMonths
Build a new siteFar more than needed18 to 24 months
The options once the gap is quantified. Ranking is by cost and lead time against a 7,200-carton-a-day shortfall.

What a 5 does differently

A 3 sizes the gap and picks the cheapest option that closes it. A 5 sizes the gap and then separates the two questions the operations director has merged. Next year's 12% shortfall is an operating problem solvable in weeks with no capital. Whether the network needs a second site is a strategic question about where growth is coming from and how far the customers are from this one building, and it should be decided on its own evidence rather than smuggled in as the answer to a throughput gap. Saying that clearly is the recommendation the CFO actually asked for.

A 5 also stress-tests the cheap options rather than banking them. Overtime at 40 minutes a day is fine as a bridge and corrosive as a permanent state, because it resurfaces as attrition and error rates. A re-slotting programme assuming a 9% rate gain should be sized against what re-slotting has delivered before, not against what the arithmetic needs. And if growth holds at 25%, the year after next needs 75,000 cartons a day, beyond every option except the new site, so the site question has a deadline even though it is not next year's answer.

One closing honesty about all five prompts. They are missing the three things that make a real case hard. There are no exhibits you have to read cold and interpret under time, because here the numbers arrive pre-digested in a list. There is no interviewer pushing back mid-answer, asking why you assumed 70% penetration, or telling you your structure has a gap and waiting. And you cannot mark your own presence: whether you sounded calm, whether you answered the question actually asked, whether you listened. Those are the parts of the scorecard a written example cannot reach, which is why worked prompts are practice rather than rehearsal.

The closest substitute for the missing parts is a spoken case. We run six full cases across industries, each with real exhibits, live pushback and a report scored on the three-dimension, fourteen-skill rubric where 3 means you met the first-round standard. The first is free, and there is a three-minute spoken taster at /try. The coaches are AI personas, built by the two of us from the way we ran interviews, and we say so plainly because you should know what you are practising against.

Common questions

How many case interview examples should I work through before a first round?

Eight to twelve properly worked cases is enough for most candidates, provided each one is done out loud, under time, and marked afterwards. Volume past that point produces diminishing returns quickly. Two cases worked to a written self-assessment beat ten cases skimmed, because the interview tests whether you can produce a structure, not whether you recognise one.

Are these real McKinsey, BCG or Bain cases?

No, and nobody publishing case examples online has real ones. The firms do not release their case libraries, and interviewers sign confidentiality agreements covering them. Every prompt in this guide is written in-house against an invented company. What transfers is the shape of the problem, the arithmetic standard and the marking criteria, all of which are consistent across firms even though the specific cases are not.

Should I read the model answer if I get stuck halfway through?

Finish first, badly if necessary. Getting to a wrong recommendation teaches you more than reading a right one, because the interview will require you to produce an answer from an incomplete picture. If you are genuinely stuck, skip forward one step only: read the next line of the working, then close it and continue on your own.

What is the difference between a case example and a practice case?

An example is a written prompt with a worked solution you compare yourself against afterwards. A practice case is spoken, timed, and has someone on the other side who asks questions you did not anticipate and pushes back on your assumptions. Examples build structure and arithmetic. Only practice cases build the composure and listening that a quarter of the scorecard measures.

How do I know whether my answer was good enough?

Score the transcript, not the feeling. Mark yourself 1 to 5 on framing, rigor and synthesis, writing one line of evidence before each number, and treat 3 as meeting the first-round standard rather than as average. If you cannot point to the specific moment that justifies the score, the score does not stand and the honest mark is lower.

Which case type should I practise first?

Profitability, then market sizing. Profitability teaches the diagnose-before-you-prescribe discipline that every other case shape borrows, and its tree is the one you can genuinely reason from first principles. Market sizing teaches audible arithmetic and sense-checking, which carry into pricing, capacity and entry economics alike. Pricing and operations are worth saving until the first two feel automatic.