At the end of a first-round day the interviewers sit in a room, and each of us gets about a minute per candidate. The minute is not a replay of your case. It is one sentence, sometimes two, and a score. "Structured, fine on the numbers, never took a position." "Drove it well, fell over when I pushed." As far as your outcome is concerned, that sentence is the interview.
Most candidates have the scoring backwards. The sheet in front of the interviewer has no row for mistakes. It has four to six rows for capabilities, and a mistake costs you only when it changes what gets written in one of those rows, or said in that minute. The slips candidates lie awake over rarely get that far. A short list of behaviors almost always does.
By the end of this piece you should be able to play back a recording of yourself, predict the sentence an interviewer would say about it, and pick the one drill that changes the sentence.
The sheet has rows for skills, not for errors
Firms do not publish their scoring forms, and I will not describe any one firm's. What they publish is thinner than prep sites pretend. As of September 2026, McKinsey's interviewing page says only that the case exists "to evaluate your analytical thinking and approach to solving complex problems" (McKinsey, Interviewing). BCG's case preparation page is more specific: structure your approach, ask thoughtful questions, analyze data, perform quick calculations, identify the most important factors, communicate clearly, and expect that there is not always a single right answer (BCG Careers, Case Interview Preparation). Bain says any feedback "will be based on an objective scoring system" (Bain, Our Hiring Process), the closest any of the three comes to admitting there is a sheet.
Fold those together with the forms I have filled in and you get the composite below. The row names are mine; the behaviors are the ones I have watched move a score.
| Row | A strong score sounds like | A weak score sounds like |
|---|---|---|
| Structure | A tree built for this client's decision, first branch to test named aloud | A generic profit tree, no view on where to start |
| Analysis | Set-up before arithmetic, units carried, result checked against a reference point | Numbers arrive without a set-up, or arrive wrong and stay wrong |
| Judgment | Every figure turned into a consequence for the client | Correct figures with nothing attached |
| Drive | You name the next step; the interviewer mostly says yes | The interviewer names every next step |
| Communication | Headline first, then support; check-ins that are real questions | Narration of everything you think, in the order you think it |
| Synthesis | Recommendation, two reasons, one risk, unprompted, before time is called | A summary of the discussion, delivered when asked |
Notice what is not on it. "Made an arithmetic error" is not a row. "Forgot a branch" is not a row. Those are evidence, and whether evidence becomes a score depends on what happens in the thirty seconds after it.
How a slip becomes a sentence
In my experience a slip reaches the debrief when at least one of three things is true.
It repeated. One dropped zero is a slip. A second dropped zero, or a second number without units, is a pattern, and a one-minute summary is made of patterns. I cannot usefully say "she made a mistake at minute eleven" in a debrief. I can say "she doesn't check her numbers," and everyone writes it down.
It went unrepaired. The interviewer almost always hands you an exit: "Does that seem about right?" "Say more about why." Take the exit and fix the slip and you have demonstrated exactly what the Analysis row rewards. Defend the slip and you have converted an arithmetic problem into a judgment problem, a different and heavier row.
It happened in the last five minutes. Interviewers finish their notes after you leave, and the close is the freshest thing in memory. A muddled recommendation colors every earlier note. A crisp one buys forgiveness for a wobble at minute eight.
Candidates tend to believe the reverse: that the verdict is in by minute two and the rest is theater. The best field evidence I know of says otherwise. Frieder, Van Iddekinge, and Raymark tracked 1,042 real job interviews run by 166 interviewers and asked each interviewer, after each one, when the decision had been made (publication record, Journal of Occupational and Organizational Psychology, 2016; BPS Research Digest summary).
| When the interviewer said the decision was made | Share of 1,042 interviews |
|---|---|
| Within the first minute | 4.9% |
| Between minute 1 and minute 5 | 25.5% |
| Between minute 5 and minute 15 | 29.5% |
| After minute 15 | 17.7% |
| Still undecided when the interview ended | 22.5% |
Two caveats. These were campus recruiting interviews, not consulting cases, so treat it as evidence about interviewers rather than about case scoring. And the "four minutes" folklore it tests probably originates, per the BPS summary, in a 1954 doctoral thesis, a thin foundation for a belief that shapes how many candidates prepare. The useful detail: interviewers who asked every candidate the same questions took longer to decide, not less, and case interviews are about as structured as interviews get. A poor first minute sets a prior. It does not settle anything.
Ned's rule. A slip you catch is a note. A slip you repeat is a sentence. A slip you defend is the verdict.
Six behaviors that reliably make the sentence
These are the ones I hear most in decision meetings, roughly in the order I hear them, each with the sentence it turns into.
| Behavior in the room | The sentence it becomes | Why it survives |
|---|---|---|
| Waiting for the interviewer to name each next step | "Couldn't drive it." | Repeats at every transition; by minute ten it is the dominant pattern |
| Solving the problem next to the one asked | "Didn't answer the question." | Cannot be repaired late; twelve minutes on the wrong question is twelve minutes gone |
| Correct numbers with no consequence attached | "Fine on the math, no judgment." | Every calculation is another chance to omit the so-what |
| Defending a number after being offered the exit | "Didn't check it, then dug in." | One slip lands in two rows, Analysis and Judgment |
| Acknowledging a nudge and carrying on regardless | "Didn't listen." | Visible, rare among strong candidates, and interviewers remember being ignored |
| Producing the recommendation only when asked | "Never landed it." | Happens in the last five minutes, and prompted synthesis scores a 3 at best |
The first row is the one candidates misdiagnose most. In a candidate-led case the interviewer's job is to answer your questions and hand you data. If you finish an analysis and stop, the silence is not thoughtful; it is a request to be told what to do next, and each time I fill it I make a small mark. Three marks and the sentence writes itself. The fix is mechanical: every result ends with "so next I'd look at X, because Y." You do not need to be right about X. You need to have said it before I did.
The second row has the cheapest fix and gets the least practice. Bain's case preparation page tells you to "clarify the objective" and make sure you understand the question before diving in (Bain, Preparing for the Case Interview). Repeat the question back in one sentence before you build anything, and write it at the top of your page so it is still there at minute twenty.
Not on the list: the canned framework opening everyone trains against. It is a weak start that the first good next step repairs, and it reaches the debrief only when passivity follows it.
Slips that almost never reach the room
The other half of the diagnosis is knowing what to stop worrying about: what candidates ask me about most, against what I actually write down.
| What you are worried about | What gets written down |
|---|---|
| An arithmetic slip you catch yourself | "Self-corrected," a positive in the Analysis row |
| Asking for thirty seconds to think | Nothing; silence with a stated purpose is not a communication failure |
| Not knowing an industry term | Nothing, if you ask; McKinsey's own page wants problem-solvers from law, medicine, engineering, or another field entirely |
| A framework that misses a branch | A note at most; I will add the branch and watch what you do with it |
| A hypothesis the data then contradicts | Nothing, if you kill it yourself when the data arrives; a wrong hypothesis dropped cleanly reads as judgment |
These are single events with a cheap repair, and the sheet is built to ignore single events. A candidate who spends the last two weeks of prep eliminating them is polishing a row that was never going to fail them.
Worked example: one dropped zero, two debriefs
Kestrel Bakehouse is an invented bakery chain: 120 stores, $1.4 million revenue per store per year, average ticket $9. Should it add delivery? The interviewer supplies fixed delivery costs of $36,000 per store per year (driver hours, packaging, platform fee), an average delivery order of $18 at a 40% contribution margin, and a 60% in-store contribution margin.
Two candidates make the same slip on the first calculation, chain revenue.
Candidate A: "120 stores at 1.4 million is 16.8 million." I ask whether that seems right for 120 bakeries. "Yes, bakeries are small businesses." We move on. Later A computes the break-even correctly and says 14 orders a day "feels like a lot for a bakery," with no reference point, because A never built one.
Candidate B: "16.8 million. No, hang on. 120 times 1.4 million is 168 million; 16.8 would be 140 thousand a store, which is a kiosk, not a bakery." Fifteen seconds. Then the break-even, then the sentence that decides the case.
Here is B's arithmetic, which you can reproduce. Contribution per delivery order: $18 × 40% = $7.20. Break-even orders per store: $36,000 ÷ $7.20 = 5,000 a year, about 14 a day over 350 trading days. Current transactions per store: $1.4 million ÷ $9 = about 155,000 a year, roughly 440 a day. So 14 delivery orders is about 3% of today's traffic, which sounds easy. But if a delivered order replaces a walk-in, the store gives up $9 × 60% = $5.40 of in-store contribution, and the incremental contribution falls to $1.80.
| Cannibalization assumption | Incremental contribution per order | Break-even orders per store per year | Per trading day | Share of today's 440 daily transactions |
|---|---|---|---|---|
| None: every delivery is new demand | $7.20 | 5,000 | 14 | 3% |
| Half of deliveries replace a walk-in | $4.50 | 8,000 | 23 | 5% |
| Every delivery replaces a walk-in | $1.80 | 20,000 | 57 | 13% |
B's close: "The answer turns on cannibalization, not demand. At zero cannibalization this is easy; at full cannibalization the bar is four times higher and I'd doubt it. I'd pilot in ten stores, measure walk-in traffic before and after, and hold the platform fee negotiation until we know which row we're in."
| Row | Candidate A | Candidate B |
|---|---|---|
| Analysis | 2: error stood, defended when offered the exit | 4: same error, caught in one sentence, check stated aloud |
| Judgment | 3: correct break-even, no reference point, no decision | 5: found the one variable the decision depends on |
| Debrief sentence | "Doesn't check her numbers and dug in when I gave her the exit." | "Slipped once, caught it herself, found the variable that matters." |
Same error, same minute. One is a note; the other is the verdict.
Hear the sentence before the interviewer says it
Run the debrief on yourself after every mock. Before any feedback, your partner gives you a verdict of twelve words or fewer, in the form "[strength], but [the row that failed]." You write yours without conferring. If they match, you have a diagnosis. If not, the gap is nearly always the same: you graded events ("I got the math wrong at minute nine") and your partner graded rows ("you never took a position"). Rows are what the real room grades, so adopt your partner's sentence.
Then drill that row instead of running another full case. A full case gives every row one rep; a drill gives the failing row fifteen.
One recording, one sentence, one drill
Run one full case today, with a live partner or the voice case with Ned, which takes about eighteen minutes and ends in a debrief scored on seven dimensions. Before you read the debrief, write your twelve-word sentence. Then open the scores: the lowest row is your sentence, whether or not you predicted it. Give the matching drill fifteen minutes:
- "Couldn't drive it": structure drills, saying the next step aloud before you read the next prompt.
- "Never landed it": synthesis drills against a 45-second timer, recommendation first.
- "Fine on the math, no judgment": math drills with one rule: no number without a "therefore."
- "Didn't check it, then dug in": quick math, reference point aloud before the answer.
- "Didn't answer the question": restate the question in one sentence before every tree, and check the recommendation against it at the end.
Never done a case? Start with three typed turns: five minutes, no account, each turn scored.
Everything on CoachNed is open for seven days, no card; after that it is $120 for a recruiting season or $49 a month. CoachNed is independent and not affiliated with McKinsey, BCG, Bain, or any other firm named here.
Frequently asked questions
What is the most common case interview mistake?
In first rounds, the sentence I hear most in debriefs is about drive: the candidate waited for the interviewer to name every next step. It outranks math errors by a wide margin because it repeats at every transition and cannot be repaired after the fact.
Will one math mistake fail a case interview?
Almost never on its own. A slip you catch, or fix when the interviewer asks whether the number seems right, is a note at most. The slip that fails you is the one you defend, because it moves from the Analysis row into the Judgment row.
Do interviewers decide in the first few minutes?
The field study above found 4.9% of decisions made in the first minute and 22.5% still open when the interview ended, and interviewers using consistent questions took longer to decide. A weak opening sets a prior you have twenty minutes to move.
Should I tell the interviewer when I notice a mistake?
Yes, in one sentence, then fix it. "I've dropped a zero; that should be 168 million" costs nothing and scores as self-correction. A long apology or a restart from scratch costs time, the one resource in the room you cannot repair.
Sources
- Interviewing at McKinsey — McKinsey's current description of the problem-solving interview and of the backgrounds it recruits from. Checked 2026-09-24.
- BCG Careers: Case Interview Preparation — what BCG says a case interview asks of you, and that there is not always a single right answer. Checked 2026-09-24.
- Bain & Company: Our Hiring Process — Bain's statement that interview feedback is based on an objective scoring system. Checked 2026-09-24.
- Bain & Company: Preparing for the Case Interview — Bain's advice to clarify the objective before diving in. Checked 2026-09-24.
- Frieder, Van Iddekinge, and Raymark, "How quickly do interviewers reach decisions?", Journal of Occupational and Organizational Psychology 89(2), 2016, publication record — bibliographic record and DOI for the study of 1,042 interviews by 166 interviewers. Checked 2026-09-24.
- BPS Research Digest: Do interviewers really make a hiring decision in the first four minutes? — summary of the same study with the decision-timing percentages and the probable 1954 origin of the "four minutes" folklore. Checked 2026-09-24.