Case Interview Prep: The Structure, Math, and Practice That Actually Transfer

A case interview hands you a business problem with no answer key — declining margins at a retail chain, whether a manufacturer should enter a new market, what to pay for an acquisition. The interviewer is not scoring whether you already know the industry. They are scoring four things: can you take an ambiguous question, break it into parts, ask for what you need, and land on a recommendation someone could act on. Everything below is aimed at those four things, and none of it requires a business degree.

The two formats you will actually meet

Prep guides describe one generic case. Real interviews come in three shapes, and the shape changes what you rehearse.

Firms change formats more often than prep guides do. Before you build a plan around one format, read the firm's own candidate materials for the specific round you booked — that page is more current than any third-party guide, including this one.

The anatomy of one case

A 30-minute case has the same five beats whether the prompt is about a coffee chain's margins or a pharma launch. Knowing the beats is most of the calm.

1. Listen and take notes — the first 2 minutes. Write the question down verbatim. Every case contains one sentence that defines what success means (restore margin, not grow revenue); find it before you do anything else.
2. Clarifying questions — 30 to 60 seconds. Two or three, maximum: objective, scope, constraints. Do not ask what the objective is when the prompt already stated it.
3. Structure out loud — 2 to 3 minutes. State your framework, then immediately name the first branch you want to open and why. The structure is a map for the next 20 minutes, not a presentation.
4. Analysis — 10 to 15 minutes. Math and exhibits, one branch at a time. Narrate what you are about to compute and why before you compute it.
5. Recommendation — 3 to 5 minutes. Answer first, then three reasons, then risks and next steps. If time runs out, skip straight here: a thinner analysis with a clear recommendation scores better than a rich analysis with no answer.

Frameworks: four worth knowing, and how to build your own

The argument against frameworks is that memorized ones produce generic answers. That is true of checklists, not of issue trees. The difference is whether the structure is built from the question or fitted to it — and interviewers can tell within a minute which one they are hearing.

Build your own the same way every time: start from the question, split into two to four branches that do not overlap, split each branch once more. Two levels deep is almost always enough; a third level is usually procrastination. If two branches could both contain the same fact, they are not a framework — they are a list.

The six calculations that cover most cases

Cases reuse the same arithmetic. Learn these six and you can compute your way through the large majority of prompts. The skill being scored is not the arithmetic — it is doing it cleanly, out loud, with a sanity check attached.

One worked example, said the way it should be said: fixed costs of $2M a year, price $50, variable cost $30. Contribution is $20 per unit, so breakeven is 100,000 units — which is $5M of revenue at that volume, and roughly a fifth of the market sized earlier in the case. That last clause is the sentence that gets scored. The bare number is not.

Reading an exhibit in 30 seconds

Most cases include a chart, a table, or both. The failure mode is reading all of it, out loud, in order. Read in this sequence instead:

• Title and units first. "Revenue by segment, $M, 2019-2024" already tells you what question the exhibit was built to answer.
• Axes and legend. Absolute or indexed? Does the scale start at zero? A truncated axis exaggerates a small change — noticing that out loud is itself scored.
• The outlier. The one bar, line, or cell that breaks the pattern. That is almost always why the exhibit is in the case.
• One sentence back to the structure. Which branch of your framework this belongs to, and whether it confirms or kills that branch. An exhibit you cannot connect to your structure is an exhibit you did not need.

The recommendation is a separate skill

Candidates lose more points in the last three minutes than in the middle fifteen. The format that works is the one consultants use with clients: answer first, support second.

Answer first, three reasons, risks, next steps. If the interviewer pushes back, that is not failure — the case is continuing. Defend what you can defend, concede what you cannot, and do it quickly; a candidate who reverses a position the moment it is questioned scores worse than one who holds a defensible position.

A practice plan that survives contact with a real case

Reading cases is not practice. Saying answers out loud to another person is. This plan fits around a job or a degree, and the ordering matters more than the totals:

1. Ten solo cases, structure only. Read the prompt, write your structure and your first question, stop. Twenty minutes each. This is where frameworks actually get built — everything else rehearses what this creates.
2. Ten cases out loud with a partner, alternating interviewer and candidate. The partner's only jobs are to interrupt with "why?" and to hand you an exhibit you did not plan for.
3. Two full mocks, timed, no pausing. The first will be bad. That is the point — it calibrates your pace, which is the thing solo practice cannot teach.
4. Keep a one-page log per case: prompt, structure you used, calculation you fumbled, question you could not answer. The fumbles repeat across cases; the log is how you stop repeating them.

Firm-specific prep is the last 10%, not the first: read the firm's own description of its format, and if it is a written or video case, practice writing answers to fixed questions under a timer, because there is no one to clarify with. Everything above still applies. For where case prep fits in the wider schedule, the interview prep time budget splits the hours into reusable and per-interview work — case reps belong in the reusable half, and the same discipline of quantifying what you did is what makes a resume line credible, covered in resume work achievements.

The mistakes that cost offers

• Opening with a framework before restating the question. You answer the prompt you expected instead of the one you got, and the interviewer hears it.
• A checklist with no connection to the prompt. The 4Ps answer to a margin case tells the interviewer you have one structure and will use it regardless.
• Math with no sanity check. A number off by 10x that nobody catches, including you, is worse than a slower correct answer.
• Silence while computing. Narrate. An interviewer cannot credit thinking they cannot hear.
• No recommendation because you ran out of time. The recommendation is the product; the analysis is the receipt.
• Treating pushback as failure. Cases are built to push. Candidates who get quiet after the first challenge lose the room.

FAQ

Do I need to memorize frameworks? Four is enough, and only as raw material. Build your structure from the prompt in front of you every time — a memorized checklist produces an answer that fits every case and therefore none of them.

How much math prep is enough? When each of the six calculations above takes you under 30 seconds out loud, including the sanity check, without a calculator. Until then, drill them daily for 15 minutes — it is the cheapest point gain in case prep.

Is it acceptable to ask clarifying questions? Yes, and you should — two or three, and they should change how you approach the case (scope, objective, constraints). Asking something the prompt already answered does the opposite of building confidence.

How long should the recommendation be? 60 to 90 seconds: the answer, three supporting reasons, the main risk, and the next step. Longer than that and you are re-running the analysis instead of concluding it.

I have never done a case. Where do I start? Ten solo structure-only reps before your first out-loud rep. Structuring is the separable skill — get it automatic first, and the live case becomes a smaller problem. Your opening 60 seconds still matter before any of this; the 60-second self-introduction is the other half of the first impression.