McKinsey Case Interview 2026: The Complete Prep Guide

What a McKinsey case interview is

A McKinsey case interview is a single conversation in which the interviewer owns the agenda. You get a prompt, you give a structure, and then the interviewer hands you one question at a time: a calculation, an exhibit to read, a list to generate, a recommendation to deliver. The business problem is real in shape and invented in detail, and it comes from the kind of work the firm sells.

Three things sit inside the same recruiting process, and candidates mix them up.

1. Solve. A gamified assessment taken online for most consulting roles, before the interviews.
2. The personal experience interview. A conversation about things you have actually done, widely shortened to the PEI.
3. The problem-solving interview. The case. Some roles add an expertise interview on top.

McKinsey publishes no length for any of them and no count of rounds. It publishes one sentence that settles both questions: "Your recruiter will confirm your interview details in advance." Read that as an instruction. The invitation in your inbox is the authority on how many conversations you have, how long each one runs, and what shape it takes.

One sentence on the firm's page invites a specific mistake. It says all candidates have at least one individual interview and that the number of colleagues you meet will vary. That sentence sits under internal roles, meaning the firm's own operations and support jobs. It describes those jobs. Build stamina for one conversation more than you expect anyway, because the marginal case is where tired candidates lose the structure they held all morning.

How the McKinsey recruiting process runs

Two stages are documented for client-facing roles, and the order between them is documented too.

1. Application, usually with Solve. The firm writes that for most consulting roles the process includes Solve, "a gamified assessment that tests your natural problem-solving abilities". Technical roles get assessments built for their skills instead.
2. Interviews. A personal experience interview, then a problem-solving interview, with expertise interviews added for some roles.

The entry title decides which seat you are being read for, and McKinsey publishes it. Its undergraduate page enters bachelor's hires as Business Analyst and describes an internship of eight to ten weeks. Its MBA page points students to Associate and Summer Associate openings, so Associate here is a graduate-school seat. Establish which ladder you are on from the posting, because the same word sits at a different height at other firms, and our BCG guide reads that firm's page in full.

Neither page publishes a bar for each level, so what follows is our reading. An analyst is read on potential, meaning raw structure and clean arithmetic under time. A post-MBA candidate is read on judgment already used, meaning a decision taken with a result someone else can check. Nothing about the case itself changes. What changes is what counts as a result in the experience half, where a university society you genuinely ran is strong evidence at analyst level and thin at post-MBA level.

The motivation question inside the experience interview has a public source. The firm lists 25 industry practices and 13 capabilities, and a specific reason for choosing McKinsey comes off that list.

The interviewer-led format, and what it changes

Interviewer-led means the interviewer picks the next step. McKinsey does not use the phrase on its pages, so treat the label as an accurate description of observed behavior rather than firm policy. The behavior is real, and it is the thing candidates adapt to worst.

In an interviewer-led case you give a prompt back, lay out a structure, and then wait. The interviewer takes your structure and asks a specific question, sometimes one that skips the branch you were about to explore. You answer that question. You do not choose the path.

In a candidate-led case, common at other firms, nothing tells you when to move on. You announce what you want to look at, you go there, and you decide when a branch is finished.

The practical adjustment is small and hard. After your structure, stop. Do not walk your own branches uninvited. Wait for the question, answer exactly that, then say what it makes you want to check next. The interviewer either follows you or redirects you, and both outcomes tell you where the case is going.

What a McKinsey case interview is scored on

McKinsey publishes the format of the case interview and publishes no marking scheme. No firm in this section publishes one. The weights below are ours. They are the dimensions our own grader measures on a structure, each out of a hundred.

The personal experience interview, theme by theme

The personal experience interview is the half McKinsey documents best, and the firm hands over the questions. Its interviewing page carries an interactive module naming four themes, and under each theme it gives its own definition and one sample question.

Two of those definitions are sharper than they look. McKinsey defines Personal Impact as respectful debate inside a trust-based relationship, and the question it prints under that theme asks for a disagreement. A candidate who arrives with a triumph has answered a different theme. Courageous Change carries the same trap: the firm defines it as learning and openness to feedback, and its question asks what you had to learn fast, which is a long way from bravery.

Keep the fourth name straight. The theme is Courageous Change. Problem Solving is what McKinsey calls the second interview, and merging the two puts a theme on your grid that the firm never published.

Filling eight PEI slots with five stories

McKinsey asks you to come prepared with two personal examples for each of the four themes. Two times four is eight slots, and nobody has eight distinct career-defining stories. About five experiences have to cover them, which means a story must work against more than one theme.

That single arithmetic fact is why the common approach fails. One rehearsed answer per rumored question collapses the moment the interviewer asks for a second example of the same theme.

Build a grid instead of a list. One row per experience, one tick per theme it can genuinely carry:

A column with fewer than two ticks is the theme you are short on, and the one to go and find an experience for. A row with one tick is a fragile story, because it survives its intended question and nothing else.

The firm also narrows what it wants inside each story. It asks you to focus on "your specific role" and to describe "the key actions that were critical to success". An answer that describes what the team achieved, or how hard the situation was, has spent its time on the two parts nobody asked about. The fit practice covers the probe that breaks a thin story.

The four practice cases McKinsey publishes free

McKinsey publishes four complete practice cases on its interviewing page and writes that "these cases contain logic, thought processes, and suggested answers". They are the closest thing to a real McKinsey case that exists in public. They open in a browser with no account and no download.

Work them in that order and you get variety rather than repetition. Two of the four are judged on something other than profit, which is exactly the shape that breaks a memorized framework. Do them aloud and timed, then read the suggested answer for the reasoning rather than the number.

A longer list of McKinsey cases circulating elsewhere came from candidates who sat the real thing. The firm's assessment integrity expectations prohibit sharing the content of an interview or assessment, including specific questions and problem-solving scenarios, so four published cases are worth more than forty leaked ones.

One case, from prompt to recommendation

Five moves, with a real number carried through.

1. Play back the objective. "A regional grocery chain has seen profit fall 18% over two years, revenue is flat, and we want to know why and what to do. Success is profit back to its prior level. Over what horizon?" Ten seconds, and it catches a misunderstanding while it is still free.
2. Ask two or three things that change the structure. Past three, you are stalling. Here: is this the whole industry or just this chain? Is the estate the same size as two years ago? Those two answers reshape everything after them.
3. Structure, without naming a framework. "Revenue is flat and profit is down, so the movement is on the cost side or in mix. I'd split cost into cost of goods and operating cost, then check whether the decline is the whole estate or a subset of stores." The branches came out of the arithmetic on the table.
4. Run the number aloud, and write the base down first. Forty stores averaging $6m is $240m of revenue. The 18% is 18% of profit, and nobody has yet said what profit was. Ask for it. At $12m two years ago, an 18% fall is about $2.2m of profit gone. Only now can a margin figure be set against it. Two points on $240m is $4.8m, more than twice the gap. A branch that explains twice the decline has a bad assumption inside it, so say so: margin cannot have fallen two full points across the whole estate.
5. Recommend, first sentence. "Measure margin store by store before cutting anything, because two points across the estate would be $4.8m against a $2.2m fall in profit. My read is that a minority of stores carries most of the loss. I would rank the estate by margin decile first. The risk of moving sooner is cutting cost in the stores that still work."

A recommendation, a supporting figure, a named risk and a next step, in four sentences. Step four is where this case is won or lost. A percentage fall in profit is a percentage of profit, and a margin movement is measured on revenue. Compare the two before the profit base is on the table and you will reach a confident answer about the wrong quantity.

The math a McKinsey case expects

The arithmetic in a McKinsey case is school arithmetic done out loud, under time, with units attached. Percentages, ratios, growth over a few years, a market size built from a population and a rate. McKinsey publishes no arithmetic standard for the case interview, so the weights below are ours, from the math rubric our grader runs.

Three habits carry most of the mark, and the worked case above shows all three.

1. Write the base down before you move. A percentage means nothing until the quantity it applies to is on the table. Asking for the profit figure is faster than reconstructing it later.
2. Say the unit every time. Dollars of revenue, dollars of profit and points of margin are three different things, and the most common failure in a case is comparing two of them.
3. Say what the number means before the interviewer asks. A figure delivered silently scores on one dimension out of four, because our math rubric puts 40 on the answer and 60 on working shown, units carried and meaning stated.

Round aggressively and announce the rounding. Forty stores at $6m is easier than thirty-eight at $6.2m, and the rounding is part of the working you are showing. If a result looks wrong, say it looks wrong and name the assumption you suspect. A candidate who catches an implausible answer has demonstrated the thing being tested.

The math drill and the structuring drill are open without an account, and both report the dimension you lost rather than a score alone.

Solve, the assessment before the interviews

Solve is the online assessment McKinsey runs for most consulting roles, before any interview. The firm describes it as a gamified assessment created to showcase your problem-solving abilities, and its line on preparation is four words long: "There's no need to prepare".

The digital assessment page answers the question most guides get wrong. It says your performance "will be considered in conjunction with the rest of your application and any other assessments you are asked to take", and its Solve FAQ answers "No" to the question of whether the game alone decides a candidacy. So it travels with the application rather than acting as a gate that disappears afterwards.

Six logistics items from the FAQ cost candidates their sitting, and most preparation guides carry none of them.

1. Your invitation email carries the length. No McKinsey page publishes one. Each task opens with an untimed tutorial before the clock starts.
2. You start and finish in one sitting. The firm's own introduction video says so on camera.
3. A technical fault you do not report cannot be fixed afterwards. The FAQ is explicit: with no contact to support at the time, there is no exception and no reset. Support costs you nothing, since the team grants extra time or resets the link depending on the fault. The address is [email protected].
4. It runs on a computer only. No tablet, no phone. A tech check runs after you click your link.
5. Four languages are available, English, Spanish, Portuguese and Japanese, with Iberian and Latin American variants of the two Romance languages.
6. You may be asked at random to take an additional task in person or to explain the logic you used.

Two lines explain how integrity is enforced. Solve uses AI methods to vary each scenario so that no two candidates get the same parameters, which makes comparing notes with last week's candidate useless. And during the assessment you may not use applications, websites, AI tools or pre-prepared notes, with exclusion, a rescinded offer or legal action as the stated consequences.

On the preparation contradiction, plainly: the firm says no prior knowledge is tested, and that is true, because Solve tests no business content. Rehearsing the format removes one cost only, which is meeting the interface under a clock for the first time. Our Solve simulators reproduce the timing and constraints and teach no content, because there is none to teach. Case practice and screen practice are separate work, and our McKinsey Solve page carries the video in full.

What McKinsey publishes about using AI

McKinsey prints a boundary on AI rather than a warning, and it is precise enough to act on.

The firm sums the boundary up in one sentence worth memorizing: it wants to get to know you rather than your AI tool.

One item on its remote interview checklist catches honest candidates. The firm asks you to "disable previously installed AI software (such as note-taking tools or virtual assistants), as these may unintentionally record or assist with the interview". A transcription bot you set up months ago for your own meetings can join a video call without being invited. Check what is installed the day before, while there is still time to remove it.

The rest of its remote checklist is ordinary and worth doing anyway: a quiet private space, headphones, charger and water within reach, camera on, virtual backgrounds and blurring switched off, and a test run of your own video feed.

Six ways candidates lose a McKinsey case

Each one maps to a dimension of the mark, which is why naming them beats a list of tips.

1. Naming a framework instead of using one. Announcing a framework spends the opening on a label. The branches are the content, and a Concision loss is the cheapest one to avoid.
2. Missing a whole branch. Coverage is 45 of 100 because a branch you miss cannot be recovered later. The one most often missed is that the cause is external.
3. Arithmetic with no working shown. A correct figure nobody can check scores on one dimension out of four.
4. Never reaching a lever. A structure that stays at the top level covers the problem and answers nothing. Depth is 20 points, and it is where "so what would you do?" lands.
5. A recommendation that arrives last. Answer first is 35 of 100. Two minutes of recap loses the room before the answer arrives.
6. Walking your own branches uninvited. You were not asked. It reads as poor listening, and in this format that costs the most.

Five of the six are habits rather than knowledge gaps, which is why reading about cases changes a score less than running one and reading the correction. The sixth, missing a branch, is the one that responds to drilling structures in volume.

How to prepare for a McKinsey case interview

There is no fixed six-week calendar here, because the right length depends on which dimension you lose. Six steps, in this order.

1. Diagnose, in twenty minutes. One short drill of each type: math, structuring, brainstorming, exhibits, synthesis, sizing. Read the dimension each rubric names as the first correction. That list is your plan, and it is usually not the plan you would have written.
2. Work the losing dimension until it changes. A full case exercises six skills at once and reports one impression, which makes it expensive for fixing one thing. Drill the single dimension, then return to full cases.
3. Build the theme grid. Five experiences, eight slots, using the table above. Any column under two ticks is a gap to close before anything else.
4. Answer the four published questions aloud, timed. They are the firm's own wording, so no better rehearsal exists. A story that reads well in notes runs three minutes in the room, and the minute past the point of interest does the damage.
5. Work the four published cases, one per sitting, then read the suggested answers for the reasoning.
6. Rehearse the screen separately, once the case mechanics hold. Different exercise, and hours on one do not transfer.

With three days rather than three months: do steps 1 and 4, plus the synthesis part of step 2. Those are the cheapest points on the board. With a month, add one full case per sitting and alternate the four published ones with cases a partner reads to you, so that somebody else picks the next question.

What McKinsey does not publish

Five things circulate about this interview that no source we could read supports, and naming them is more useful than guessing at them.

The number of rounds and cases. Not published, on any of the documents above. The firm writes instead that your recruiter confirms your interview details.

The length of an interview, and the length of Solve. Neither is published, and for Solve the FAQ says why: it is in your invitation email. Any fixed duration circulating elsewhere came from somewhere other than the firm.

The share of candidates who reach the final round. Not published, by the firm or by anyone with access to its data. Rates circulate, and none of the ones we found named a source that can be opened.

The Solve pass threshold, and the number of tasks. The firm's video names Red Rock Island and the Sea Wolf and walks through both, then says the assessment has "multiple interactive, game-based tasks" without ever giving a total.

Whether an expertise interview applies to you. The firm says some roles add one, and never says which. Your recruiter knows, and it is worth asking before you plan, because a coding exercise and a case do not share a single hour of practice.

How this page is sourced, and what changed

Every statement above about McKinsey is either a quotation from one of its own pages or an absence named as one. The readings we add on top are marked as ours: how the entry title moves the bar, and what our grader weighs. McKinsey's servers refuse connections from our network, so each page was read either through a text reading proxy or in a dated Internet Archive snapshot of the firm's own page, and the method sits next to every source in the register at the foot of this page. If your own experience of this process contradicts something here, write to us with what you saw and when.

8 October 2026. The shape of the page. It was built around the four free cases, and the cases are now one section of a full preparation guide, with the format, the scoring, the arithmetic and the preparation order given their own.

7 October 2026. What Solve's tasks are called. An earlier version said the task names in circulation appear on none of the McKinsey documents we had read. The firm's own introduction video names Red Rock Island and the Sea Wolf in its English captions and walks through both tasks. We read that caption file on 7 October 2026.

29 September 2026. The name of the fourth experience theme. An earlier version called it Problem Solving. The module McKinsey embeds on its interviewing page names it Courageous Change, and Problem Solving is what the firm calls the second interview. The two had been merged, which put a theme on the preparation grid that does not exist.

29 September 2026. The arithmetic in the worked case. The old step four set a margin movement measured on revenue against a profit fall measured on profit, with no profit base ever stated. Step four now asks for the profit figure first.