Somewhere around the third minute of a case, the interviewer makes a decision that never appears on the scoring sheet: whether to hand you the case or run it for you. Only the first earns a strong score on structure.
That decision is made on your opening, and most guides prepare you for the wrong test: a framework presentation, judged on completeness. From the other side of the table it is a bid for control. The interviewer has a case pack that was fixed before you walked in, and while you talk they are matching your branches against it. Name the places its analyses live, in a form they can act on, and they say "where would you like to start?" Miss them, and they say "let's look at costs first," and note that they had to.
You will leave with a word budget for the opening, a test for whether a branch is real, and a worked opening to score yourself against.
What the interviewer is doing while you talk
In the candidate-led cases I ran, I had a pack: a prompt, three to five exhibits, and a page of numbers I released when asked the right question. My job during your opening was to work out how much of it you would find on your own.
McKinsey's notes on its own sample case tell candidates to "mention a range of potential issues to explore instead of immediately diving very deep into one issue," then to "ask your interviewer if he or she wants to go deeper on any of them specifically" (McKinsey, GlobaPharm sample case). That second sentence is the hand-off written down. You propose the map, the interviewer chooses the route, and how far they must redirect you is the information they are collecting.
So the opening is scored twice: formally, under whatever the sheet calls structure, and informally, as a running count of steers. The count is what gets said in the debrief: "needed a lot of prompting to get to costs." I do not remember a candidate who recovered a top structure score after three steers in the first ten minutes.
Early judgments stick. In a study of 189 accounting students in mock interviews, interviewers' first impressions after a two-to-three-minute rapport-building stage correlated at r = .42 with the scores they gave after the structured interview that followed, and at r = .22 with the internship offers students later received from real firms (Barrick, Swider, and Stewart, 2010). Modest correlations, formed before a single job-relevant question, and driven by judged competence rather than liking. Your opening is a far richer sample than small talk.
Three minutes, and a third of it is silence
The clock below is what good candidates did. Conversational English runs at roughly 164 words a minute measured turn by turn (Yuan, Liberman, and Cieri, 2006). A candidate pointing at a page and choosing words runs slower; my working figure is 140 a minute, Ned's estimate rather than a measurement.
| Element | Clock | Your words | What the interviewer is checking |
|---|---|---|---|
| Recap, with a number in it | 0:00–0:20 | 35–50 | Did you hear the decision and the yardstick |
| Two clarifying questions, with answers | 0:20–0:50 | 35–50 | Do they change the work |
| Silence while you build the page | 0:50–1:50 | 0 | Nothing. They are reading the pack |
| Tree, three or four branches | 1:50–2:40 | 100–140 | Does it map onto the exhibits |
| Where you want to start, and why | 2:40–3:00 | 30–45 | Will they have to steer |
Roughly 200 to 280 spoken words across three minutes. At 400 you are not thorough, you are unmappable.
The four-branch ceiling is not folklore. Working memory holds about four chunks when rehearsal is blocked (Cowan, 2001), and your interviewer is also holding the pack, the clock, and their next question. A fifth branch is the one they will remember as padding.
Candidates fear the silence most. Bain's interview-prep page lists "Take a moment: Pause to gather your thoughts when needed" among its case tips (Bain, preparing for the case interview). A minute of silence after "may I take a moment to structure this?" costs nothing. A minute of talking while you think costs the hand-off.
The recap should contain a number
The recap is the cheapest twenty seconds in the case, and the most common early leak.
It proves you heard three things: who the client is, what decision they need, and what number they will judge it by. Use the interviewer's words for the decision. If they said "the fund wants to know why margin fell," do not upgrade it to "help them transform the business." Add a goal they did not state and they stop trusting your listening.
Then do the thing almost nobody does: turn the numbers in the prompt into one derived number. A revenue figure and two margins give you the profit that went missing. One multiplication, said aloud, inside the recap. It shows quantitative instinct before any exhibit arrives, and it sizes the prize.
Ask only what changes the tree
Two clarifying questions is the right number; a third changes the interviewer's face. The filter: if the answer cannot change your first branch, do not ask it.
| You want to know | Ask or assume | Why |
|---|---|---|
| The yardstick (margin, cash, share, wait time) | Ask | Every branch is measured in it |
| Time horizon or a hard constraint (no layoffs, exit in 18 months) | Ask | Kills or creates whole branches |
| A fact that was in the prompt | Never | Tells them you did not listen |
| Whether the market is growing, whether costs rose | Never | That is the case. You are asking for the answer |
| Anything you can reasonably assume | State the assumption | "I will assume US only unless you say otherwise" earns the point without spending a question |
A correction to a stated assumption is not a steer; it is a fact you obtained in four seconds. Phrasing is covered in the clarifying-questions guide.
Branches the interviewer can act on
Ned's rule. A branch is only a branch if I can picture the exhibit that answers it. If I cannot, you have named a topic, and topics do not get you the wheel.
"Costs" is a topic. "Which cost line moved, labor or supplies; I would want the cost structure for both years" is a branch, because I can picture the exhibit and check whether I have it. When all four branches match my pack, I hand you the case.
Two consequences. MECE is table stakes, not the goal: the interviewer has no exhibits for most of the universe, so spend the silent minute making each branch answerable rather than perfecting the partition. Tree-building mechanics are in the issue tree guide.
And state a lean, not a hypothesis. Bain's recruiting blog asks candidates to "make hypotheses about which factors are most important and take a position supported by clear reasoning" (Bain, how to stand out in the recruiting process). "I believe this is a cost problem" is a guess. "I would start with clinic mix, because six points of margin on flat revenue is too large to be pricing" is the same lean with its reasoning attached, and the reasoning earns the point.
Worked opening: Marlowe Vet Group
An invented case; the numbers are mine.
The prompt. "A private-equity fund owns Marlowe Vet Group, a chain of 40 veterinary clinics in the US Midwest. Revenue has been flat at about $180 million for two years, but EBITDA margin has fallen from 18 percent to 12 percent. The fund plans to sell in 18 months and wants to know what happened and what to do about it."
Recap, 47 words. "So Marlowe is a 40-clinic vet chain with flat revenue of $180 million, and margin has dropped six points in two years. That is roughly $11 million of annual EBITDA gone on the same revenue, and the fund needs it back, or explained, before an exit in 18 months."
Six percent of $180 million is $10.8 million. "Roughly $11 million" is the whole point.
Two clarifiers, 45 words. "Two questions before I structure. Were any of the 40 clinics acquired in the last two years, or is this the same estate throughout? And is the fund's yardstick margin percentage at exit, or EBITDA dollars, in case a smaller, more profitable footprint is acceptable?"
Answers: twelve clinics were acquired 20 months ago, and the fund cares about EBITDA dollars.
Silence, 60 seconds. Write the tree.
Tree, 108 words. "I would look at four things. First, where the margin went: is the drop spread across all 40 clinics or concentrated in the twelve acquired ones? I would want margin by clinic cohort for both years. Second, which cost lines moved: vet and technician labor, drugs and supplies, rent, and corporate overhead, as a share of revenue, then and now. Third, whether flat revenue is hiding a demand problem: visits and average ticket separately, because price rises can mask falling volume. Fourth, what is fixable in 18 months: scheduling and procurement move fast, leases do not, so this branch decides what the fund can actually promise a buyer."
Lean, 44 words. "I would start with margin by cohort. If most of the six points sits in the acquired clinics, this is an integration problem with a known playbook. If it is everywhere, it is a labor-cost problem, which is harder to fix in 18 months."
Total spoken: 244 words. Every branch names its exhibit, and the lean hands the interviewer an easy next line: "Good. Here is margin by cohort."
The same case, opened the way I heard it most often:
"I would like to use a profitability framework. Profit is revenue minus cost, so I would look at revenue, broken into price and volume, and at costs, broken into fixed and variable. I would also want to understand the market, the competition, and the customer, and then look at any internal issues at the company."
Fifty-five words, no silence. My scoring of the two, not any firm's rubric:
| What I am listening for (1 to 5) | Framework | Marlowe |
|---|---|---|
| Heard the decision and the yardstick | 2 | 5 |
| Branches I can map to an exhibit | 1 | 5 |
| Quantitative instinct before any data | 1 | 4 |
| Reasoning attached to a lean | 1 | 4 |
| Steers I expect to need | 3 or more | 0 or 1 |
The framework version is not wrong. It maps to nothing in my pack, so I would have to drive.
When the interviewer is driving from the start
Interviewer-led cases, McKinsey's format among them, change the delivery, not the tree. McKinsey describes the aim as understanding "how you structure tough, ambiguous business challenges, identify important issues, deal with all the implications of facts and data, formulate conclusions and recommendations, and articulate your thoughts in a fast-moving discussion" (McKinsey, interviewing).
So: recap in one sentence, still with the number. Deliver the same tree in half the words and end with the move from the McKinsey note quoted earlier, offer the range, then ask where they want to go deeper. The steering count still exists here. It measures how often their next question was one you had already named.
Practice this today
Take ten case prompts. For each, set a three-minute timer and open out loud: recap with a derived number, two clarifiers (answer them yourself), a silent minute, a three-or-four-branch tree, a lean with a reason. Record it. Count the words; over 300, cut branches or adjectives, never the silence. Then apply the exhibit test to each branch.
CoachNed's structure drill runs this loop without a full case.
When ten openings feel routine, sit one full live voice case. UrbanBrew runs about 18 minutes and the debrief scores structure separately from math and synthesis, so you learn whether the opening is your leak. For a five-minute first rep with no account, use /start. Everything is open for seven days, no card; then $120 for a recruiting season or $49 a month.
CoachNed is independent and not affiliated with McKinsey, Bain, BCG, or any other firm named here.
Frequently asked questions
How long should the opening of a case interview take?
About three minutes at candidate-led firms, including up to a minute of silence, and closer to ninety seconds in interviewer-led formats. Count words, not seconds: 200 to 280 stays mappable.
Should I ask for time to think before giving my structure?
Two clarifying questions is the right number; a third changes the interviewer's face. The filter: if the answer cannot change your first branch, do not ask it.
Yes, and say so: "May I take a minute to structure this?" Bain lists pausing among its case tips, and McKinsey's sample-case notes say taking time to organize your thoughts "tells the interviewer that you think about the problem in a logical way."?
In the candidate-led cases I ran, I had a pack: a prompt, three to five exhibits, and a page of numbers I released when asked the right question. My job during your opening was to work out how much of it you would find on your own.
How many clarifying questions should I ask at the start of a case?
Two, occasionally three, each of which changes a branch of your tree. State everything else as an assumption or park it, and never ask the question the case exists to answer.
What if my structure misses what the interviewer wanted?
Fold it in out loud: "That changes my second branch; I would go straight to the cost lines." Updating on a new fact is scored well; pretending the tree already covered it is scored badly, and interviewers can tell.
Sources
- McKinsey & Company, Interviewing — what the problem-solving interview assesses. Checked 2026-09-24.
- McKinsey & Company, GlobaPharm sample case — notes on organizing thoughts and offering a range of issues. Checked 2026-09-24.
- Bain & Company, Preparing for the case interview — case tips, including pausing. Checked 2026-09-24.
- Bain & Company, How to stand out in Bain's recruiting process — recruiter guidance on frameworks and supported positions. Checked 2026-09-24.
- Barrick, Swider, and Stewart, Journal of Applied Psychology, 2010 — first impressions versus later interview scores (r = .42) and offers (r = .22). Checked 2026-09-24.
- Cowan, Behavioral and Brain Sciences, 2001 — working memory holds about four chunks. Checked 2026-09-24.
- Yuan, Liberman, and Cieri, Interspeech 2006 — conversational speaking rate of about 164 words a minute. Checked 2026-09-24.