Of the eight candidates on a typical first-round day, two or three will visibly freeze at some point: mid-tree, mid-multiplication, or right after the interviewer says "I'm not sure I agree." I have never marked anyone down for the freeze. I mark down what happens in the ten seconds after it, and that part is trainable in a way that "staying calm" is not.
Pressure does not make you slower or stupider; it makes you smaller. It shrinks working memory, the scratchpad you use to hold a question, three numbers, and a so-what at once, and it shrinks it most in the people who normally have the most of it. So the candidate who "doesn't test well" is very often the strong one who never learned to put anything on paper. The fix is not a lower heart rate. It is needing less working memory at the moment the heart rate arrives.
By the end you will know what the sheet actually records when you wobble, how to run a math step on two slots instead of nine, which pre-room routines have evidence behind them, and one drill for tonight.
What pressure takes from you, specifically
The cleanest study I know on this is Beilock and Carr's 2005 experiment in Psychological Science, "When high-powered people fail". People solved arithmetic problems under low and high pressure. Only the participants with high working-memory capacity got worse, and only on the problems that made the heaviest demands on working memory. The low-capacity group, who the authors suggest were already relying on simpler strategies, did not decline. Their conclusion: pressure "harms individuals most qualified to succeed by consuming the working memory capacity that they rely on for their superior performance."
How much capacity is there to consume? Nelson Cowan's 2001 review in Behavioral and Brain Sciences put the real limit of short-term storage at about four chunks, not Miller's famous seven. Four, on a good day. Now subtract whatever the worry is occupying: "he looked at his watch", "was that 6.5 or 6.8", "if I fail this I have no offers".
That is the whole mechanism. You did not forget how to multiply. You ran out of slots while multiplying, and the slots you lost were holding the inputs.
The candidates who froze in my rooms were rarely the weak ones. They had the best transcripts, had done everything in their heads for twenty years, and had never needed a method that survived losing two slots. Nerves are a working-memory tax, and the tax hits the biggest earners.
What the sheet records, and what it cannot
Scoring sheets differ by firm, and I will not pretend to know any firm's current confidential rubric. The ones I have used shared a shape: dimensions such as structure, analytics, judgment, and communication, plus a free-text box. There is no heart-rate box and no "seemed nervous" dimension. What ends up in the free-text box is behavior, and behavior is where stress becomes a score.
| What you feel | What the interviewer observes | What I write down | Cost |
|---|---|---|---|
| Heart pounding, dry mouth | Nothing | Nothing | None |
| Twelve-second blank, labeled | A deliberate pause | Often nothing; sometimes "good control" | None, occasionally a plus |
| Twelve-second blank, unlabeled | A lost thread | "Lost it after the exhibit" | Communication minus |
| Arithmetic wobble, caught | A self-correction | "Caught own error" | Neutral to plus |
| Arithmetic wobble, not caught | A wrong number delivered with confidence | "Math wrong, didn't check" | Analytics minus, sometimes decisive |
| Recommendation ends with a question mark | No recommendation | "Needed prompting for a rec" | Judgment minus, usually decisive |
The costly rows are not the ones where you felt worst. They are the ones where something that belonged on the page stayed in your head.
McKinsey's own interviewing page calls the case a "problem-solving conversation" and offers two pieces of advice: practice the sample cases, because that will "boost your confidence when navigating the problem-solving conversation", and "relax and be yourself". Only one of those is actionable, and it is the one about reps.
Ned's rule. I have never failed a candidate for a pause I could name. I have failed plenty for a number I could not check. When you must choose between looking composed and writing it down, write it down.
One math step, counted in slots
A fictional case. Kestrel Coldchain runs 320 refrigerated trucks. Each does about 95,000 miles a year at 6.5 miles per gallon, and diesel costs $3.80 a gallon. A retrofit would lift fuel economy to 7.2 mpg. The interviewer asks for the annual fuel saving.
Count what a candidate doing this in their head is holding: the question (saving, annual, dollars), the formula, five inputs (trucks, miles, two mpg figures, price), the running total, and the so-what. Nine live items against a capacity of roughly four. Something gets dropped, usually an input, which is how a bright candidate ends up saying "$1.4 million?" with a rising intonation after forty seconds of silence.
The same step, run so that no line needs more than two live items:
| Line on the page | What you say | Items in your head |
|---|---|---|
| Saving = today's fuel spend x (1 - 6.5 / 7.2) | "I'll get today's fuel bill, then the share we save." | 0 |
| Miles = 320 x 95,000 = 30.4M | "320 times 100,000 is 32 million, less 1.6 million." | 2 |
| Gallons = 30.4M / 6.5 = about 4.7M | "6.5 times 4.5 is about 29, so roughly 4.7 million gallons." | 2 |
| Spend = 4.7M x $3.80 = about $17.9M | "Just under 18 million dollars of fuel." | 2 |
| Share saved = 0.7 / 7.2 = 9.7%, round to 10% | "The retrofit saves about a tenth of the fuel." | 1 |
| Saving = 10% x $17.9M, trimmed for rounding | "Roughly 1.7 to 1.8 million a year." | 1 |
Exact answer: 4.677 million gallons, $17.77 million of fuel, 9.72 percent saved, $1.73 million. The rounded path lands within a few percent, and every intermediate number is on the page where we can both check it. An interviewer who spots a slip in visible working usually points at the line rather than writing "math wrong". When it is in your head, they have nothing to point at.
The two candidates here have identical arithmetic ability. One needed nine slots and had four. The other needed two and had four.
In the room: spend words, not slots
Every in-room move that works under stress does the same thing: it takes an item out of your head and puts it on the page or into a sentence. Calming down is a side effect, not the mechanism.
| Moment | The move | The sentence |
|---|---|---|
| You go blank building the structure | Label the pause, write three nouns, then speak | "Give me twenty seconds to lay this out properly." |
| Numbers arrive faster than you can hold them | Write the equation with blanks before any digit | "Let me set this up before I compute." |
| The interviewer pushes back | Write the objection as one line, then answer the line | "So the challenge is that competitors could match the price. Let me take that." |
| You notice you are speeding up | Speed is your body racing the overflow; put the pen on the next line and read from it | "Coming back to the tree: branch two was cost." |
A labeled pause of twenty seconds feels like a minute to you and like nothing to me; I am usually glad of the chance to write. A candidate who re-reads their tree aloud after a wobble is literally reloading working memory from the page. It looks like recovery because it is.
Before the room: what the evidence supports
Most nerves folklore is untested. Three interventions are not.
| Intervention | What the study found | Sample and setting | What I would do with it |
|---|---|---|---|
| Expressive writing | Test-anxious students who wrote about their worries for ten minutes right before a final scored 6 percent higher than anxious controls, a B+ against a B-. Writing about unrelated topics did nothing. | Two lab studies and two ninth-grade biology finals; Ramirez and Beilock, Science, 2011 | Ten minutes in the lobby. Write what you are afraid of, in sentences. Nobody reads it. |
| Reappraising arousal | Participants told that arousal helps performance scored 739 against 684 on a practice GRE quantitative section, and 770 against 706 on the real GRE one to three months later. No effect on verbal. | 60 students, 28 with follow-up scores; Jamieson and colleagues, 2010 | One line when the symptoms start: "this is my body getting ready." Small study; cheap and plausible rather than proven. |
| Structured breathing | Five minutes a day of cyclic sighing (two inhales, one long exhale) for 28 days improved mood and lowered breathing rate more than mindfulness meditation. Heart rate and heart-rate variability did not change. | 108 adults, mostly Stanford undergraduates, remote; Balban and colleagues, Cell Reports Medicine, 2023 | A daily habit in the month before interviews, if you like it. Not evidence that one breath at the door changes anything. |
The two interventions with measured effects on test scores are about worry, not physiology. They free slots.
The 72 hours before: sleep and caffeine, with numbers
The most expensive thing candidates do in interview week is treat sleep as the flexible budget line. Van Dongen and colleagues restricted healthy adults to four or six hours in bed a night for two weeks and found cumulative cognitive deficits "equivalent to up to 2 nights of total sleep deprivation". The finding that should worry you: participants' own sleepiness ratings barely moved after the first couple of days. They were impaired and did not feel it. Three midnight cases on Tuesday, Wednesday, and Thursday cost you working memory on Friday, and you will walk in feeling fine.
On caffeine, the FDA's guidance for most healthy adults is that 400 milligrams a day, roughly two to three 12-ounce cups of coffee, is not generally associated with negative effects. The interview-day problem is rarely caffeine; it is a different dose than usual, in either direction, on an empty stomach. Drink what you drink on a normal Tuesday.
The routine that follows: your last full case 48 hours out, one twenty-minute math drill the day before and then nothing, a normal dinner, a normal breakfast, normal caffeine, and ten minutes of writing in the lobby. No "final review" that ends at 1 a.m., and no warm-up case on the morning; it fills the memory you need empty.
Folklore I would drop
- "Tell the interviewer you're nervous; it humanizes you." It costs twenty seconds and puts the word "nervous" in the free-text box. If you must, one clause, then start the recap.
- "Power pose in the bathroom first." The 2010 study behind this has not held up, and its first author, Dana Carney, has stated publicly that she does not believe power-pose effects are real. Stand however you like.
- "Nerves mean you're not ready." Beilock's data say almost the opposite: pressure hurts the most capable candidates most. Nerves mean you have something to lose. So does everyone else in the waiting room.
If what you have is clinical, with panic attacks or insomnia that predates interview season, that is healthcare, not a casing gap, and nothing here replaces a doctor.
Train the reload, not the calm
Tonight, run one case under a single rule: no number may be spoken before it has been written. Not the inputs, not the intermediate figures, not the answer. Catch yourself about to say an unwritten figure, stop, write it, then say it. It will feel slow for ten minutes and then like having a second brain on the desk, which is roughly what it is.
Then sit a voice case so you can hear what your own wobble sounds like. CoachNed's live voice case with Ned runs about eighteen minutes and ends with a seven-score debrief that includes communication, so a freeze you did not notice will show up in the transcript. With five minutes rather than twenty, the first rep is three typed turns of a case with instant scores and needs no account. For the math discipline, the math drill is where to practice writing the equation before the digits, and the quick-math drill is free for as long as you want it.
Everything on CoachNed is open for seven days with 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
How do I stop my mind going blank in a case interview?
You cannot reliably stop the blank; you can make it cheap. Label it ("give me twenty seconds"), write three words, and re-read your structure aloud. The blank is a working-memory overflow, and the page is the extra memory.
Do interviewers mark you down for being nervous?
Not directly. No scoring sheet I have used has a "nervous" dimension. Interviewers score the behaviors nerves cause: unlabeled silences, unchecked math, recommendations that end in a question mark. All three are fixable with method rather than composure.
Is it okay to ask for a moment to think in a case interview?
Yes. A named pause of fifteen to twenty seconds usually reads as control. What reads badly is the unnamed pause, or the third pause in five minutes. Use the time to write, not just to think.
How much sleep do I need the night before a case interview?
Your normal amount, and the same the night before that. In the sleep research, six hours a night for a run of nights produced deficits participants could not feel, which is the trap of "one more case" at midnight.
Sources
- When high-powered people fail: working memory and "choking under pressure" in math (Beilock and Carr, Psychological Science, 2005) — pressure consumes working memory and harms high-capacity individuals most. Checked 2026-09-24.
- The magical number 4 in short-term memory (Cowan, Behavioral and Brain Sciences, 2001) — the roughly four-chunk capacity figure. Checked 2026-09-24.
- Writing about testing worries boosts exam performance in the classroom (Ramirez and Beilock, Science, 2011) — ten minutes of expressive writing; 6 percent gain for anxious students. Checked 2026-09-24.
- Turning the knots in your stomach into bows (Jamieson, Mendes, Blackstock, and Schmader, 2010) — GRE quantitative scores of 739 vs 684 (practice) and 770 vs 706 (actual). Checked 2026-09-24.
- Brief structured respiration practices enhance mood and reduce physiological arousal (Balban and colleagues, Cell Reports Medicine, 2023) — daily cyclic sighing for a month; mood and respiratory-rate effects, no heart-rate change. Checked 2026-09-24.
- The cumulative cost of additional wakefulness (Van Dongen and colleagues, Sleep, 2003) — six hours a night or less for 14 days; deficits equivalent to up to two nights without sleep, largely unfelt. Checked 2026-09-24.
- Spilling the beans: how much caffeine is too much? (U.S. FDA) — the 400 milligrams a day figure. Checked 2026-09-24.
- Interviewing at McKinsey (McKinsey careers page) — the firm's "problem-solving conversation" framing and its advice to practice. Checked 2026-09-24.
- My position on "power poses" (Dana Carney, UC Berkeley Haas) — the first author's statement that power-pose effects are not real. Checked 2026-09-24.