What to take away
- Four to six weeks at five to eight hours a week is a realistic plan, and twenty to forty focused hours beats a hundred unfocused ones
- Analytical thinking is half the mark, so structuring, case maths and a decisive recommendation deserve half your preparation time
- Improvement is per-skill: drill the one skill scoring lowest rather than working on all five at once
- Peers are free but cannot hold the bar, human coaches mark best and cost most, AI mocks give unlimited reps and consistent marking but no human unpredictability
- Count marked cases, not cases: an unmarked case is a rehearsal, not practice
- The final week is a taper, not a cram: one case a day, normal sleep, one handwritten page naming your weakest skill
What is a case interview, and why do firms still run them?
A case interview is a compressed client problem. An interviewer hands you a situation a real team once spent months on, strips it down to the few numbers that matter, and asks you to get somewhere useful inside twenty or thirty minutes of conversation. The company is usually invented or anonymised. The situation is not: a client with a decision to make, data that is incomplete on purpose, and a partner who wants a view by Friday.
Firms run cases because the job is unusually hard to sample any other way. A CV tells them where you have been. A behavioural interview tells them how you work with people. Neither tells them what you do in the first hour of an unfamiliar problem in an industry you knew nothing about last week, which is roughly what a consultant does every few months for the length of a career. The case is a job simulation, and it is deliberately uncomfortable, because the job is sometimes uncomfortable too.
Read that way, the list of things an interviewer is sampling for gets short and concrete:
- Whether you can break an unfamiliar problem into parts that do not overlap, and then say which part you would open first
- Whether you can do arithmetic out loud, under mild time pressure, without losing the thread of the argument
- Whether you know which number matters commercially, and notice when an answer is implausible
- Whether a client would believe you: whether the recommendation arrives before the reasoning, and lands inside a minute
- Whether you are pleasant to think with, which sounds like a soft criterion and is not
Formats differ between firms. McKinsey rounds lean interviewer-led: the interviewer drives, poses a sequence of questions, and hands you exhibits when they choose. BCG and Bain lean candidate-led: you set the agenda and ask for the data you want. The distinction changes the choreography, not the marking. Both sample the same skills, and a candidate who has only ever practised one format gets a mild shock in the first two minutes of the other. Practise both if you are interviewing at more than one firm.
What are you actually being assessed on?
Nothing in preparation matters more than knowing what is on the scorecard, because the scorecard decides where your hours should go. Interviewers do not form one overall impression and write it up afterwards. They complete a grid: three weighted dimensions, split into fourteen named skills, each marked from 1 to 5 with a line of evidence attached.
| Dimension | Weight | The skills inside it |
|---|---|---|
| Analytical thinking | 50% | Framing, commercial judgement, rigor, creativity, synthesis |
| Presence & communication | 25% | Presence, precision, listening, relationship-building |
| Track record & drive | 25% | Self-direction, collaboration, depth, growth mindset, results delivered |
Fourteen skills sounds unmanageable until you notice that five of them carry half the total. Framing, commercial judgement, rigor, creativity and synthesis are worth as much between them as everything else on the page combined. Track record and drive is marked from your behavioural stories rather than from the case itself, which is why a case-only practice session cannot score it at all.
The 1-to-5 scale is criterion-referenced, and that is the part most candidates misread. A 3 does not mean average, it means you met the first-round standard. A 1 does not mean bad, it means the skill never appeared: the candidate who attempts the maths and fumbles scores above the candidate who asks to be walked through it. Weighted and averaged, 4.5 and above reads as outstanding, 3.75 and above as strong, 3.0 and above as solid. We treat 3.5 as the readiness line, because a real first round wants you comfortably above the standard on every skill rather than exactly on it.
Two consequences for planning. First, improvement is per-skill rather than general: you do not get better at cases, you get better at synthesis, or at rigor, one cell at a time. Second, the cheapest points available to anybody are usually on their weakest skill, not their strongest. Because the overall is an average, closing a 2 moves the number further than turning a 4 into a 5. Our companion guide on scoring sets out every anchor, cell by cell, and it is worth reading before you build a plan.
How long does case interview preparation really take?
Four to six weeks, if you can find five to eight hours a week and you spend them on the right things. Shorter is workable for candidates who already do quantitative work for a living. Longer usually means months of drifting, because this material decays quickly when it is not concentrated: a structuring habit built in March is gone by June.
Week one is where most plans go wrong, in one specific way. Candidates spend it collecting material: four books, two courses, a spreadsheet of frameworks. Collecting feels like progress and produces almost none. One good case book, read once, is enough input for the whole six weeks. Everything after that is repetition against feedback.
| Week | Focus | Honest hours |
|---|---|---|
| 1 | Read one case book end to end. Learn the scoring dimensions. Write your first three structures on paper with no time limit | 5 to 6 |
| 2 | Case maths for fifteen minutes daily: percentage change, growth over several years, breakeven, sizing. Two half-cases, structure and maths only | 6 to 8 |
| 3 | Three full timed cases, each one marked against the rubric. Expect them to go badly, and mark them anyway | 6 to 8 |
| 4 | Fix the two lowest-scoring skills specifically. Start behavioural stories if your round includes them | 6 to 8 |
| 5 | Mixed practice: alternate interviewer-led and candidate-led, unfamiliar industries, exhibits you have not seen before | 5 to 6 |
| 6 | Taper. One case a day at most, re-read your own scored reports, rehearse openings and closings | 3 to 4 |
Add the column up and the plan comes to roughly thirty to forty hours. That is the whole thing. We say it plainly because the opposite claim, that you need a hundred cases, does real damage: twenty to forty focused hours beats a hundred unfocused ones, and the hundred-case candidates we see are usually a hundred repetitions into the same comfortable case shape, slightly quicker at it than they were. A focused hour has a named target skill, produces a score, and ends with one thing written down to fix. An unfocused hour is a case you did because the evening was free.
- A named target skill decided before you start, not identified afterwards
- A timer, because untimed practice trains a habit you cannot use on the day
- A transcript or a recording, because you cannot mark a memory
- One written fix at the end, carried forward into the next session
Two adjustments to the plan are legitimate. If your first round is in ten days, do not attempt six weeks compressed: run the week 2 maths drills daily, do four full timed cases, mark them, and leave the reading list alone. If you have three months, do not stretch the plan to fill them. Run it in the last six weeks of your window and use the earlier time on the behavioural block, which is the part almost everybody leaves until it is too late.
Which skills should you build first?
Five skills carry the case. The order below follows the weighting and it follows what we see in the transcripts we mark: work down it, and do not start the next one until the one above is holding.
Framing, which is not framework recall
On the bar looks like three or four branches that fit this specific question, do not overlap, and cover the ground, followed by a sentence naming which branch you would open and why. A memorised profitability tree bolted onto a market-entry question scores below three home-made branches that address what was actually asked. The drill is not learning more frameworks: it is building structures for twenty different prompts on paper, ninety seconds each, reusing nothing.
Case maths, out loud and auditable
On the bar means arithmetic that is right, or right in order of magnitude, with assumptions said aloud and a sense-check at the end. Silent maths is unscoreable even when the answer is correct, because the interviewer marks the chain and not the number. Fifteen minutes a day for three weeks does far more here than one long weekend, and four shapes cover most of what appears: percentage change, growth over several years, breakeven, and sizing from a population.
Commercial judgement, the hardest to shortcut
On the bar means knowing which branch matters before you have the data, knowing what a number implies for the client, and noticing when a result is too good to be true. This skill grows slowest, and the only reliable way to build it is exposure: read what the firms publish, read the business pages, and after every case ask yourself what you would have told the client if you had been handed one number instead of six.
Synthesis, the sixty seconds that decide the mark
On the bar means recommendation first, then two or three reasons carrying numbers, then the largest risk, then the next step, inside a minute. Hedging costs more than being wrong: a reasoned view that turns out incorrect outscores "there are several factors and I would want more analysis". Rehearse this apart from full cases. Take a case you have already done and deliver the close ten times until it fits in sixty seconds without notes.
Communication, which you fix last and cheaply
On the bar means answering the question actually asked, in numbers when numbers were asked for, using the interviewer's nudges instead of talking through them, and staying calm under pushback. It sits last because it is the fastest to move once the first four are in place, and because polishing delivery around a weak structure is the most common way candidates waste a final fortnight.
Framing and synthesis are first and last on this list for the same reason they are first and last in the room. They are the opening ninety seconds and the closing sixty, they are what the interviewer remembers most clearly at the debrief, and a failure in either caps the score no matter how good the middle of the case was.
How should you practise, and what does each method cost you?
There are five realistic ways to practise, and none of them is best on every axis. The honest way to choose is to name what each one actually gives you and then combine two or three.
| Method | Cost | Realism | Feedback you get |
|---|---|---|---|
| Solo reading | A book or two | Low: no clock, no interruptions, no pressure to speak | None, and it feels productive, which is the trap |
| Solo drills | Low | Medium for maths and structuring, nil for conversation | Self-marked, or marked by a tool if the drill has an answer key |
| Peer mocks | Free | High: a real person, real awkwardness, real time pressure | Generous and unreliable, because your peer does not hold the bar |
| Paid human coaching | The highest, per hour | Highest, if the coach interviewed for a firm recently | The best available: specific, calibrated, priced accordingly |
| AI mock interviews | Low per session | High on the case, lower on human unpredictability | Per-skill and consistent, marked the same way every time |
Availability is the axis people forget. A coach who was interviewing at Bain last year is worth what they charge and can usually see you twice before your round. A peer is free and can practise nightly. That makes the pair complementary rather than competing: use peers for volume, and spend one or two coaching sessions late in your preparation rather than in week one, when almost any feedback would have helped and the expensive kind is wasted.
The honest limit of peer mocks is marking, not effort. Your friend has read the same books you have. They will tell you it went well, they will not catch that two of your branches overlapped, and they cannot place your synthesis against the standard because they have never watched a hundred candidates attempt it. That is not a criticism of your friend. It is a structural problem with peers marking peers, and no amount of goodwill fixes it.
Our own product sits in the last row, and we will be exact about what it is. The coaches are AI personas, built by ex-MBB interviewers, and they are never presented as human beings. What that buys you is unlimited reps at any hour and marking that applies the same rubric to every transcript, which is the one thing a rotating cast of peers cannot do. What it does not buy you is a human interviewer: an AI will not shift in its chair when it loses you, and reading a real person's discomfort is genuinely part of the job. Treat it as the volume and calibration layer. If you can also get in front of a human before your round, do that.
In practice that means a first case is free, the default session is a thirty-minute case-only run, and a sixty-minute full interview including the behavioural block exists for when you want the whole thing scored. Around it sit an eight-module course for the method and a hundred short drills at /drills for the fifteen-minutes-a-day habit, which is the part of the plan that is easiest to skip and hardest to fake later.
What should you do in the final week and on the day itself?
The final week is not a preparation week. It is a taper, and treating it as a last chance to cram is the most common self-inflicted wound we see. By seven days out, your structuring habit and your arithmetic are broadly what they are going to be. What can still move is sleep, calm, and the handful of sentences you have decided to say at the start and at the end.
What to do in the last seven days
- One case a day, at most. Five cases on a Saturday teaches you nothing and costs you Sunday
- Re-read your own scored reports rather than new material, because your weaknesses are already written down
- Write one page by hand naming your single weakest skill and the one behaviour that fixes it. That page is the only thing you read on the morning
- Rehearse your openings and closings out loud, since those are the two moments an interviewer recalls most clearly
- Keep your normal sleep schedule from three days out, and stop practising after nine in the evening
What to stop doing
- Stop collecting frameworks. A structure learned on Thursday will not be available to you on Monday under pressure
- Stop measuring yourself against other candidates' claims about how many cases they got through
- Stop practising with anyone who leaves you feeling worse without telling you anything specific
On the day
The mechanics are small and they matter more than they ought to. Headphones with a microphone rather than laptop speakers, if the round is remote. Water within reach, because a dry mouth in minute twenty reads as nerves. Two pens and blank paper turned sideways, so a structure fits across the page instead of running off the bottom of it. A closed door, and a phone in another room. None of this wins you points. All of it removes ways to lose them.
Then there is the opening. The first sixty seconds set the interviewer's expectation for the next thirty minutes, and they are almost entirely rehearsable: confirm the objective, ask for the time you need, and commit to a structure out loud.
“Thank you, that is helpful. Before I start, may I check one thing: is the client's objective the profit figure specifically, or the margin?”
“Could I take about forty-five seconds to lay out how I want to approach this?”
“Right. I would like to look at three things, and I would open with the second one, because that is where the margin gap most plausibly sits.”
Common questions
How many practice cases do I need before a first round?
Fewer than the internet claims. Eight to fifteen full timed cases is enough for most first rounds, provided each one is marked against a rubric and produces one written fix. Volume without marking plateaus fast: candidates who have done sixty cases are usually sixty repetitions into the same comfortable case shape. Count marked cases, not cases.
Can I prepare for a case interview in one week?
Partly. In seven days you cannot build commercial judgement, but you can lift the two things that cap most scores: fifteen minutes of case maths every day, and four full timed cases with a written score after each one. Skip the reading list entirely and rehearse your closing until it fits in sixty seconds. That is a defensible week, and it beats a panicked reading marathon.
Is it possible to prepare for case interviews alone?
Yes, with one caveat. Structuring, case maths and synthesis are all drillable alone, and self-marking against a written rubric works if you score the transcript rather than the feeling. What you cannot manufacture alone is unpredictability: someone interrupting you, pushing back, or handing over an exhibit you did not expect. Cover that with peer mocks, a coach, or an AI mock interview, whichever you can actually get.
Do I need a paid case interview coach?
Not necessarily, but one or two sessions late in your preparation are the best-value coaching money buys. A recent interviewer will tell you in twenty minutes where you genuinely sit, which no peer can do reliably. Book them for week four or five rather than week one, when almost any feedback would have helped and the expensive kind is largely wasted.
How is preparing for McKinsey different from BCG and Bain?
Format, not criteria. McKinsey rounds lean interviewer-led, so the interviewer drives the question sequence and releases exhibits when they choose. BCG and Bain lean candidate-led, so you set the agenda and ask for the data you want. The scoring dimensions are close enough across all three that one preparation plan serves all three. Practise both formats if you are interviewing at more than one firm.
Where can I try a case interview question for free?
We run a free three-minute spoken taster at mbbready.app/try with no signup: one question, answered out loud, and a read on how you sound under mild pressure. A full scored case is free the first time you run one, and there is a complete sample report at mbbready.app/sample-report if you want to see what the marking looks like before committing an evening to it.