Every prep guide tells you to think out loud. One of the few firms to put the phrase in writing attached a purpose clause that almost nobody quotes: BCG's interview brochure says to "think out loud" so that the interviewer understands your direction. Direction, not process. Candidates keep the verb and drop the purpose, and what comes out is rambling, which is what thinking sounds like when it is broadcast raw.
Here is the view from my side of the table. I am writing while you talk. Roughly every thirty seconds I want to put one short line on my sheet about where you are and what you have decided. If I can, you are communicating. If I cannot, I have stopped scoring your thinking and started scoring your talking, and the second score has never got anyone hired.
Below: the test, a way to say math an interviewer can follow, and one close delivered twice so you can watch the sheet change when only the packaging does.
The instruction has a second half
BCG's careers page lists four things to do in a case, and "communicate clearly" is one of them. Its older interview preparation brochure (November 2018) is more specific: adapt your hypotheses as information arrives, and "think out loud" so the interviewer understands your direction. Bain's case interview page says the same thing in different words: "Think aloud: share your reasoning and calculations, even if you're unsure," and, for virtual interviews, "clearly describe your structure."
Read those together and the job of your voice is narrow. It tells me three things, repeatedly: which question you are on, what you have concluded so far, and what you will do next. Reasoning, yes. Calculations, yes. The interior monologue that produced them, no.
Direction and process sound similar in the room and score completely differently. "I'm going to the cost side, because labor is the only line that moved" is direction. "So, um, I'm kind of thinking about whether it's cost or revenue, it could be either" is process. Same candidate, same brain, one of them scores.
The 30-second test
A study of 2,438 recorded American English telephone conversations measured about 196 words a minute across the whole call, 236 with silences stripped out, and about 164 counted turn by turn. Call it 160 to 200 words a minute for a candidate who is not racing. Thirty seconds of you talking is 80 to 100 words, roughly one paragraph of this article.
My notes on a case have a resolution of one line per thought. "Good tree, three buckets, cost first." "Misread exhibit 2, self-corrected." That is the entire record I will have of you an hour later when I am arguing your file with two colleagues. So the test is simple: every 80 to 100 words you speak should end in something I can write in one line. In a first-round day of eight candidates, my estimate is that five fail it at least once in the first ten minutes and two never pass it. The two are rarely the least intelligent people in the building; they are the ones who took the instruction literally.
Ned's rule. If I cannot write one line about where you are every thirty seconds, I stop scoring your thinking and start scoring your talking. Nobody has ever been hired on the second score.
| Moment in the case | The line I want to write | What rambling leaves me |
|---|---|---|
| After the framework | "Three buckets, cost first, said why" | "Six buckets, no priority, three minutes" |
| After a math block | "Setup stated, $7m, 10% of revenue, so-what" | "Muttered, got a number, moved on" |
| After the close | "Answer, three reasons, risk, next step, 45 seconds" | "Summary of everything, no call" |
Four slots, and then I start dropping things
Working memory is smaller than the frameworks candidates build. Cowan's 2001 review in Behavioral and Brain Sciences puts the limit at about four chunks, three to five for most people, when items cannot be grouped into something larger.
I am a person, so this applies to me while you talk. Six unlabeled buckets with three sub-points each is eighteen items in ninety seconds. I will retain four of them, and not the four you wanted. Three consequences:
- Count before you list. "Three things drive this" gives me three labeled slots before you fill them. Without the count, each new item competes with the last for the same slot.
- Three or four buckets, named as a phrase. "Chair utilization." "Labor cost per hour." "Price and mix." Sub-points stay on your page until I ask.
- Re-label when you move. "That closes utilization. Labor next." Four words, and I never have to ask where you are, a question I hold against you.
Numbers and meaning want different speeds
Most candidates rush the figures and dawdle on the meaning. It should be the other way around.
For figures, the relevant research is on what speech scientists call clear speech. Picheny, Durlach, and Braida found that when talkers deliberately speak clearly the rate drops substantially, through pauses inserted between words and longer individual sounds; their clear-speech talkers ran at roughly half the conversational rate. Say numbers that way: not hesitantly, but with air around the figures, "seven ... million ... ten percent of revenue," so I can write them down.
Here is a math block from an original case, Tidewater Dental: 40 clinics, about $70 million of revenue, EBITDA margin down from 18% to 13% in two years. The interviewer has just said chair utilization fell from 78% to 71%.
Setup, at conversational speed: "I want to know what seven points of utilization is worth in revenue. Chairs, times hours, times the utilization gap, times revenue per chair-hour."
Arithmetic, at clear-speech speed:
| Step | Calculation | Result |
|---|---|---|
| Capacity | 40 clinics × 5 chairs × 2,000 hours a year | 400,000 chair-hours |
| Hours used today | 400,000 × 71% | 284,000 chair-hours |
| Revenue per used chair-hour | $70m ÷ 284,000 | about $246, call it $250 |
| Hours lost to the 7-point drop | 400,000 × 7% | 28,000 chair-hours |
| Revenue at stake | 28,000 × $250 | about $7m |
| As a share of revenue | $7m ÷ $70m | 10% |
Meaning, back at conversational speed, one sentence: "Ten percent of revenue is sitting in empty chairs, which is bigger than the entire five-point margin gap, so utilization is where I go first, and I would not touch price until I know why the chairs emptied."
Under two minutes, one shape: setup, units, number, so-what. The setup lets me stop you before you run the wrong calculation; the so-what is the only part that touches the judgment line on my sheet. Skip it and you have done the math for free.
The same close, delivered twice
Below is the Tidewater recommendation from two candidates who have seen identical data: the utilization drop, clinical labor up from 24% to 27% of revenue, and patient rebooking down from 62% to 54%. Read both aloud and time them.
Candidate A, 93 words, about 35 seconds:
"So, looking at everything, there's quite a lot going on. Revenue has been fairly flat, labor costs went up, which is obviously a problem, utilization went down from 78 to 71, and the recall rate dropped too, so patients aren't coming back, which could be a lot of things, maybe competition or the front desk. So overall I think it's probably a combination of cost and revenue factors, and I would recommend looking at both, maybe reducing labor costs and improving utilization through marketing, and pricing could potentially be looked at as well."
Candidate B, 120 words, about 45 seconds:
"Tidewater should fix rebooking before it touches cost. Three reasons. First, the margin drop is mostly idle chairs: utilization fell seven points, which is about $7 million of revenue on a $70 million base, more than the whole five-point gap. Second, the idle chairs trace to recall. Rebooking fell from 62 to 54 percent, so the patients exist and are leaving through the back door. Third, labor is up three points of revenue, but cutting hygienists while chairs sit empty makes utilization worse, not better. The risk is that recall is falling because of a new competitor rather than the front desk, so I would check churn by clinic first. Next step: a rebooking pilot in the five worst clinics."
Same facts. Firms do not publish their scoring sheets; these four lines are the ones I have seen in some form on every sheet I have used, scored one to five.
| Sheet line | Candidate A | Candidate B | What made the difference |
|---|---|---|---|
| Communication | 2 | 5 | B opened with the answer and a count. A opened with "a lot going on." |
| Synthesis | 2 | 4 | B ranked the causes and linked them. A listed them. |
| Business judgment | 2 | 4 | B made a call and named its risk. A recommended "looking at both." |
| Structure | 3 | 4 | "Three reasons" is retrievable an hour later. A's order was the exhibits' order. |
Communication is scored on one line and leaks into three. Delivery is not a separate skill graded alongside the analysis; it is the only channel through which the analysis reaches the sheet.
The budget behind Candidate B is reproducible. At 160 words a minute, 45 seconds is 120 words: the answer in 15, three reasons at 25 each, risk and next step in 30. Anything that does not fit is a fourth reason to cut or a sentence that belonged earlier in the case.
Silence, with a label on it
Bain's tips page says outright to "take a moment: pause to gather your thoughts when needed." I have never marked a candidate down for a labeled silence. I have marked many down for the noise they made to avoid one.
The label is what makes it work. Unlabeled silence reads as stuck. "Give me 45 seconds to lay out the structure" reads as control, and buys you the 45 seconds. The going rates: 30 to 60 seconds to build a framework, 20 to 40 to run arithmetic once the setup is spoken, 10 to 15 on first sight of an exhibit. Two moments call for talking immediately: a direct question, which gets the answer before the reasons, and a blank, which gets "Let me restate the question. It was X. I'm returning to Y."
The candidates who do this well give a duration and beat it. Saying "thirty seconds" and taking twenty-five is a small promise kept, and people deciding whether to put you in front of a client notice small promises.
Practice this today: a 45-second close
Take any case you have already done. Set a phone timer for 45 seconds and record yourself closing it: the answer in one sentence, "three reasons," the reasons, one risk, one next step. Count the words. Over 130, cut a reason. Under 90, you skipped the risk or the next step. Do it three times on the same case; the third attempt is usually the first that would score a 4.
Then run it against something that scores you. The synthesis drill hands you a case's worth of findings and grades the close you type. The live voice case is the real test: about eighteen minutes of UrbanBrew by voice, then a seven-score debrief in which communication and synthesis sit on separate lines. If you have never done a case, start with three typed turns, which needs no account.
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 is not affiliated with BCG, Bain, McKinsey, or any other firm mentioned here.
Frequently asked questions
How much should you talk in a case interview?
Less than you think, at a higher density. A 30-minute case has you speaking for perhaps 15 to 18 minutes, which at conversational rates is 2,500 to 3,500 words; the interviewer keeps about one line of notes per 80 to 100 of them, so the useful measure is lines produced, not minutes talked.
Is it okay to be silent in a case interview?
Yes, if you label it and give a duration. Bain's own interview tips tell candidates to pause to gather their thoughts, and twenty-five seconds of labeled quiet scores better than twenty-five seconds of hedging.
How do you think out loud without rambling?
Broadcast direction, not process: which question you are on, what you have concluded, and what you will do next. Record a case and pause it every 30 seconds; if you cannot summarize the last stretch in one line, neither could the interviewer.
Should you say "does that make sense" in a case interview?
Once, after the framework, when you actually want the answer. Asked after every paragraph it reads as a request for reassurance, and the interviewer starts wondering whether you would ask a client the same thing.
Sources
- Case Interview Preparation | BCG Careers — BCG's public list of what to do in a case, including "communicate clearly." Checked 2026-09-24.
- BCG Interview Preparation brochure (PDF, November 2018) — "think out loud" so that the interviewer understands your direction; "overall effective communication" as something the case assesses. Checked 2026-09-24.
- Case interview preparation | Bain & Company — Bain's tips to think aloud, share reasoning and calculations, pause when needed, and describe your structure clearly. Checked 2026-09-24.
- Yuan, Liberman, and Cieri, "Towards an Integrated Understanding of Speaking Rate in Conversation," Interspeech 2006 — conversational rates of 196 words a minute overall, 236 net of silences, 164 turn by turn, from 2,438 telephone conversations. Checked 2026-09-24.
- Cowan, "The magical number 4 in short-term memory," Behavioral and Brain Sciences, 2001 — working-memory capacity of about four chunks. Checked 2026-09-24.
- Picheny, Durlach, and Braida, "Speaking clearly for the hard of hearing II," Journal of Speech and Hearing Research, 1986 — clear speech is substantially slower than conversational speech, through pauses between words and longer sounds. Checked 2026-09-24.