McKinsey Case Interview Questions and Answers: All 6 Question Types

McKinsey cases are scripted. The interviewer has the case materials, a set sequence of questions, and target answers for each. Your job is not to run the problem end to end. It is to answer the question in front of you, then take the next one.

This is the opposite of BCG and Bain, where you drive: you set the structure, choose the next analysis, and ask for data. Candidates who prepare only candidate-led cases walk into McKinsey and keep trying to grab the wheel. It reads as not listening.

Interviewer-led has three practical consequences. First, transitions are abrupt. The interviewer may close a discussion you were enjoying and hand you an exhibit. That is pacing, not a verdict. Second, you cannot hide. If your case math is shaky, a scripted math question is coming anyway. In a candidate-led case you might route around it; here you will not. Third, scoring is per question. Each answer is marked on its own, so one weak answer does not sink the interview, and one brilliant framework does not carry it.

Prepare accordingly. Drill each question type until none of them is a weak spot, because the format guarantees you will face all of them.

The write-up after your interview covers four things. Structuring: can you break a messy problem into parts that cover it, tailored to this client rather than recited from a memorized tree. Quantitative reasoning: can you set up the right calculation, run it without a calculator, and keep the units straight. Judgment: when a number lands, do you know what it means for the client, and do your ideas survive contact with how the industry works. Communication: can you speak top-down, lead with the answer, and summarize under pressure.

Every question in the case maps to one or two of these. The opening structure question is scored on structuring and communication. The math question is scored on quant and judgment, because the follow-up is always what the number means. Brainstorming is scored on structuring and judgment. The final synthesis is scored on communication above all.

Two things follow. You do not need to be spectacular on every dimension; you need no red flags on any. And interviewers compare notes within a round, so a weakness that shows up twice gets believed. If you know your weakest dimension, that is where your preparation hours go.

Take the time. Say "give me a minute to lay this out," write for 60 to 90 seconds, and come back with something you can defend. Nobody wins the case in the first ten seconds and plenty lose it there.

A defensible structure here has two properties. It covers the ground: margin is revenue minus cost, so both sides appear, split into parts you can investigate. And it is tailored: it names grocery-specific drivers, not generic ones. Revenue side: traffic, basket size, price and product mix, the shift toward lower-margin channels like delivery. Cost side: cost of goods and supplier terms, labor, shrink, the fixed-cost drag of underperforming stores. External frame: discounters and delivery platforms compressing the whole sector.

Present it top-down. State the branches first, then walk each with its sub-drivers, then say where you would start and why. "I'd start with product mix, because a two-point margin decline on stable revenue usually means mix or input costs, and mix is faster to check" is the sentence that separates a structure from a list.

The tailoring is what gets scored. A memorized profitability tree earns a pass. Grocery economics earn a distinctive.

This probe means your structure missed something the case script cares about. It is also asked when your structure was fine, to test breadth. Either way the response is the same: produce new categories, not new words for old ones.

The reliable technique is to change axis. If your structure was internal, go external: competitor pricing moves, input-cost inflation, a demand shift toward eating out. If you covered ongoing operations, go one-time and structural: a new distribution center still ramping, lease renewals, an acquisition being digested. If you covered economics, go strategic: deliberate price investment to defend share, a loyalty program running ahead of its payback.

Say the axis out loud. "So far I've looked at ongoing operations. Two structural things could also do this" shows the interviewer you have a system for finding what you missed, which is worth more than the specific items.

What kills candidates here is repackaging. Offering "pricing pressure" after your structure already had "price and mix" signals you cannot hear the difference, and the probe gets repeated until the silence gets uncomfortable.

Commit. The scored behavior is a choice with a reason, and almost any reasonable choice passes. Hedging does not.

Good reasons come in three kinds. Size: start where the money is; in grocery, cost of goods runs about 70 to 75% of revenue, so a small COGS shift swamps everything else. Speed: start with what one data pull settles, like margin by category across the two years. Signal: start where the case facts already point; if the interviewer said revenue held while margin fell, the revenue branches just got less interesting.

Then name the data like a consultant in week one: a margin bridge by year, category mix shift, supplier cost indices against shelf-price moves, labor hours per store. Specific, pullable, and tied to the branch you chose.

The interviewer may push back: the client cannot get that data, what else? That is a test of flexibility, not a rejection. Offer the proxy. If category-level margin is unavailable, basket composition and supplier price letters get you most of the way there.

Structure before arithmetic. Say the equation first: annual home-charger revenue equals new EVs sold per year, times the share that get a home charger installed, times hardware and installation price, plus a replacement stream from the installed base. Once the interviewer nods, fill in numbers.

Use round numbers and label them as assumptions. US new-vehicle sales run about 15 million a year; call EV share 10%, so 1.5 million EVs. Say 60% of buyers have a garage or driveway and install a charger: 900,000 units. Hardware plus installation around $1,500. That is roughly $1.35 billion a year; call it $1.5 billion with replacements and upgrades.

Then sanity-check unprompted. Cross-check the flow against the stock: 5 million EVs already on US roads at 60% home charging implies 3 million installed units, so 900,000 new installs a year grows the base by about a third, which fits a market where EV sales are still climbing. An answer that implied the base doubled every year would not.

The interviewer does not care whether EV share is 8% or 12%. They care that the equation was right, the arithmetic was clean, and you knew which assumption moves the answer most. Close with that: the number is most sensitive to EV sales share, so that is the assumption to pressure-test.

Do not fold and do not defend. Both are scored as failures of the same skill: treating an estimate as an identity instead of a model.

Walk the chain. "My number came from 1.5 million EV sales, 60% home-install share, and $1,500 per install. To be half, one of those is high. Install share is the softest: apartment dwellers, workplace charging, and buyers who stick with a standard outlet could take it from 60% toward 30%. That alone closes the gap." You have shown which assumption bends, by how much, and why that one rather than the others.

Check the arithmetic too, briefly and without drama. A dropped zero found by you costs little. Found by the interviewer, it costs a lot.

What this probe tests is whether you can be wrong gracefully and usefully. Consultants are wrong in front of clients weekly; the job is updating in the open, showing the reasoning, and keeping the room's confidence. Candidates who negotiate with the interviewer over whose number is right have answered the real question, badly.

Set up before you compute. Say the shape: this is payback, investment over annual saving. The saving is 20% of $600,000, which is $120,000 a year. $150,000 up front against $120,000 a year is payback in a year and a quarter. Announcing the method first means a slip in arithmetic costs you little, because the interviewer can see the machine works.

Then answer the question asked, which was should, not what is the payback. A 15-month payback on hardware that lasts five years is strong; over the life of the kiosks the client nets roughly $450,000 per store. So: yes on the economics, with two checks before a 2,000-store rollout. Does shrink rise with self-checkout, since theft and mis-scans climb without a cashier? And does checkout labor come out of the cost base, or does it shift to floor roles, in which case the saving is paper. Name them in a sentence each; do not detour.

Keep units visible the whole way: dollars per store per year. McKinsey math errors are mostly unit errors, per-store against chain-wide, monthly against annual. Saying units out loud feels pedantic and prevents the error that ends candidacies.

Buy 20 seconds: "let me take a moment with this." Read the title, both axes, the units, and the footnote, in that order. Exhibit traps live in exactly those places: an axis that starts at 90 rather than zero, bars in percent while the question is in dollars, a footnote excluding the largest segment.

Lead with the insight, not the description. "Sales rose in every region" is reading the chart to the person who made it. "Growth is concentrated in the two regions where we have the fewest stores, so the network is in the wrong place" is a takeaway. The formula: headline, driver, implication for the client's question.

Then quantify one thing. Pull the two numbers that make the point concrete: the Southeast grew 12% while the core Northeast grew 2%. Exhibits are handed out precisely so you will do arithmetic on them; an answer with no numbers in it reads as chart description.

Close by connecting to the case. "If this holds, the margin problem is partly a footprint problem, and I would want store-level economics by region next." One sentence. The interviewer decides where the case goes, but that sentence shows you know where it should.

The so-what. At McKinsey a number is not an answer; a number compared to something is an answer. The silence after a correct calculation is the most common scripted moment in the format, and candidates fill it by rechecking arithmetic that was already right.

Compare the number to whatever makes it mean something. A market growing 3% means one thing if the client grows 6% and another if the case said 1%. A $450,000 per-store saving means one thing against a $2 million refit and another against $150,000 kiosks. Benchmark, then implication: "that is double the market rate, so the client is taking share, and the margin question becomes whether they are buying it with price."

If the case has given you no benchmark, say what you would compare against and take a view anyway. Against a typical grocery operating margin of 2 to 4%, a two-point swing is existential, not incremental. An imperfect benchmark beats a naked number.

This is also the cheapest scoring dimension to improve. Structuring takes weeks of drills. Ending every calculation with "which means" is a habit you can install in an afternoon, and it moves the judgment score on every math question in the case.

"As many as you can" is a trap if you take it literally. Scattered volume scores worse than structured volume. Take 20 seconds, pick buckets, and announce them: "I'll split this into changing price perception, changing actual prices, and moves beyond price."

Then fill each bucket, eight or more ideas total. Perception: a price-match guarantee on a visible basket of items, repositioning the private label, prominent unit pricing, an entry-price tier in every category. Actual price: targeted cuts on the 100 items shoppers price-check, a loyalty price tier funded by supplier terms, pack-size changes where the category tolerates them. Beyond price: a basic delivery tier, partnerships that put offers where lapsed shoppers already shop.

Range matters: some ideas cheap and immediate, some structural. Practicality matters more than novelty. "A loyalty tier funded by renegotiated supplier terms" beats "a gamified shopping app" because it shows you know where grocery margin comes from.

Finish by flagging one. "If I had to pick, the price-match guarantee, because it changes perception across the whole store for the cost of matching a few hundred items." Interviewers remember the candidates who chose.

Recommendation first. Not the journey, not "we looked at several areas," the answer: "You should pilot self-checkout in 200 stores this year. Three reasons." Then the reasons, each one a fact from the case with a number attached: payback under 15 months on checkout labor alone, the margin decline concentrated in labor-heavy urban stores, category mix stable so the labor lever is the live one. Then the risk and the next step: "The open risk is shrink, so the pilot measures it store by store before the chain-wide decision."

That structure, answer, numbered support, risk, next step, is the house style for every McKinsey deliverable. The synthesis question checks whether it is native to you or whether you default to chronology under pressure.

Two details separate strong syntheses. Numbers survive into the summary; a synthesis without figures sounds like opinion. And the recommendation is conditional but committed. "Pilot, then decide" is a real answer that respects what you do not yet know. "It depends" is not an answer.

Practice this cold. Mid-case, at a random moment, force the 60-second version. If you cannot produce it on demand, you have analysis, not an answer.

Pick the two or three analyses that would most change the recommendation, not a list of everything unexamined. This question scores judgment about uncertainty: do you know which of your assumptions is load-bearing?

Tie each item to the decision it would change. Store-level shrink data from the pilot regions, because if shrink eats more than a third of the labor saving, the rollout case collapses. A category-level margin bridge, because the structure assumed mix is driving the decline and the case never proved it. A read on how the two discounters price the 100 most-checked items, because every win-back idea assumed the client sits 5 to 10% above them.

The shape is data, then the decision that moves. An analysis that would not change the recommendation is not worth the two weeks, and saying that about one tempting analysis is an advanced move interviewers notice.

Keep it to three. The candidate who lists nine follow-ups is displaying coverage. The candidate who picks three and defends the choice is displaying judgment, and judgment is the dimension this question exists to score.