Of the closing recommendations I have scored, roughly one in three fails before the second sentence, and not because of delivery. The candidate answers a question the client never asked. The prompt was "should we acquire this company at $400 million"; the close is "the target is attractive." One in three is Ned's estimate from years of scoring sheets, not a firm's statistic.
The thesis: the close is scored as an answer to the question read out in minute one, and by minute 25 most candidates have drifted to a different question, usually the one their last exhibit was about. Structure, math, and charts leak points slowly. Synthesis leaks them in one sentence.
By the end you will be able to turn any prompt into a first sentence that passes the echo test, budget the close to about 150 words, and score your own attempt the way an interviewer does.
What the interviewer is doing while you talk
With four minutes left, most of the score is penciled in. The close is the last data point, and for a borderline candidate it settles the sheet. In my own scoring, a decided close moved a marginal candidate up a notch more often than any other minute of the case; a recap never moved anyone up.
"The CEO has just walked in, what do you tell her" is a request for the decision the case was pointed at, and the firms say so in their own materials. BCG's interview preparation brochure lists "synthesize your thoughts" among the four abilities the case tests, and the last of its case "dos" is to "go back to the original question" before recommending. Bain's published FashionCo case grades the close in two tiers: a good answer is one sentence with a number (Option B, "projected to generate $50M more" than Option A); a better answer adds risks and next steps. McKinsey's GlobaPharm practice case states the engagement in one line: "Our overall question today, therefore, is 'Should GlobaPharm acquire BioFuture?'"
Answer-first is not a prep-site convention; it is the writing standard of the people scoring you. Barbara Minto built it into McKinsey's internal writing manual between 1963 and 1973 and published it in 1985 as The Pyramid Principle; McKinsey's alumni profile says its lessons are still taught inside the firm. The US Army's Army Regulation 25-50 names "putting the main point at the beginning of the correspondence (bottom line up front)" as one of two essential requirements of writing that is "understood by the reader in a single rapid reading." Lead with the tour and end with the answer, and you are speaking to a trained reader in the wrong order.
I cannot tell you what any firm's current scoring sheet says. I can tell you what the interviewer listens for in your first ten seconds: does this sentence answer the one I read out 25 minutes ago?
Three ways the question drifts
Drift happens to candidates who did the analysis well, because 25 minutes of analysis pulls attention toward whatever was analyzed last.
From decision to description. The prompt asked whether to acquire, enter, launch, or reprice. The close says the market is attractive, the target is strong, the economics are tight. Adjectives. The interviewer wanted a verb. "Attractive" is a finding; "acquire, but at no more than $320 million" is an answer.
From the whole to the last piece. The question was how to restore margin to 12%. The last twelve minutes were a cost exhibit, so the close is a cost recommendation. The last exhibit always wants the microphone.
From this client to any client. "Companies in this space need to invest in digital." If your first sentence survives swapping in a competitor's name, it is not about the client.
The fix for all three is mechanical.
The echo test
Before you speak, find the prompt sentence in your notes; Bain's case interview guidance opens with "Clarify the objective: make sure you understand the question before diving in." Your first sentence reuses the prompt's verb and object, then adds the condition, if there is one.
Ned's rule. Your first sentence must contain the verb from the prompt. If the client asked whether to acquire, sentence one contains "acquire." If it does not, I score everything that follows as a summary, however good the numbers in it are.
| The prompt asked | Sentence one must contain | Example |
|---|---|---|
| Should we do X? | Yes or no, then the condition | "No, not at the asking price. At $320 million or below, yes." |
| Option A or B? | The option, and the margin it wins by | "B. It clears about $6 million more over three years; A only wins if renewals double." |
| Why has Y fallen, and what do we do? | The cause, then the action | "Margin fell four points because mix moved to wholesale, not because costs rose. Fix wholesale pricing first." |
| How do we reach target Z? | Whether Z is reachable, and the main lever | "Fifty million by 2028 is reachable but not organically: trend gets you to 38; the rest needs one acquisition." |
| How big is the market? | The number, and the assumption that swings it most | "About $1.2 billion a year. At 10% penetration instead of 15%, it is $800 million." |
Every example carries a number by the second sentence; the number turns a preference into a decision.
Budget 150 words, not 60 seconds
In the Switchboard telephone corpus, Yuan, Liberman, and Cieri measured conversational English at about 196 words per minute across whole conversations and 164 counted turn by turn. Close slower, at 130 to 150, because the interviewer is writing while you speak and because slower reads as considered. Sixty seconds is therefore about 150 words. Past 160 on paper, it runs past a minute aloud and is almost certainly carrying a recap.
| Beat | Words | Seconds at 150 wpm | What it must contain |
|---|---|---|---|
| Decision | 15–25 | 6–10 | The prompt's verb, a yes or no or the option, one condition |
| Reason one | 30–40 | 12–16 | The number that carries the call |
| Reason two | 25–35 | 10–14 | A second fact; directional is fine |
| Risk | 20–30 | 8–12 | The assumption that would flip you, and how you would know |
| Next step | 20–30 | 8–12 | Something with an owner, a date, or a threshold |
Over budget, cut reason two first, then adjectives. Never the decision, never the risk: one reason and a real risk beats three reasons and no risk, because the risk is where judgment shows.
Worked example: Northwind Laundry, two closes scored
The case is invented for this guide. Northwind runs 40 laundromats across Ohio and Michigan and has piloted app-based pickup and delivery of wash-and-fold in five stores for a year. The CEO's question, read out in minute one: "Should we roll it out to all 40 stores next year?"
Averaged across the five pilots, each store added $60,000 of wash-and-fold revenue against $45,000 of operating cost, lost $7,000 of walk-in contribution to cannibalization, and netted about $8,000 a year against a $25,000 launch cost: a three-year payback, with break-even at about 10 orders a day.
Then the split the candidate asked for. The two pilot stores in dense apartment ZIP codes ran about 20 orders a day, the three suburban stores about six. Urban store: $100,000 revenue, $60,000 cost, $7,000 cannibalization, net about $33,000, paid back in nine months. Suburban store: $33,000 revenue, $35,000 cost, $7,000 cannibalization, net about minus $9,000. Two at 33 and three at minus 9 average to $7,800: the blend hid two businesses. Fourteen of the 40 stores are urban, so 14 stores cost $350,000 up front and return about $462,000 a year; all 40 cost $1 million and return about $228,000, a payback of more than four years.
The summary close, about 95 words. "So, to summarize: the pilot ran in five stores for a year. Revenue per store went up by about $60,000 and costs by about $45,000, so it added around $15,000 per store. Cannibalization took about $7,000 of that, so the net is about $8,000, and with a $25,000 launch cost the payback is roughly three years. Urban stores did much better than suburban ones, 20 orders a day versus six. So overall the service is promising, especially in urban areas, but the economics are tighter than they first appear and there are risks to look into."
The answer close, about 135 words. "No, not to all 40. Roll it out to the 14 urban stores next year and hold the 26 suburban ones. Two reasons. First, the blended pilot hides two businesses: an urban store nets about $33,000 a year on a $25,000 launch, paid back in nine months; a suburban store loses about $9,000 a year after cannibalization. Second, a 40-store launch is $1 million up front for roughly $230,000 a year, a four-year payback on a model we have run for twelve months. The risk is that 'urban' is a read from two stores. Break-even is about 10 orders a day, so I would set the kill line at 12. Next step: rank the 14 by apartment density, launch seven in spring and seven in fall, and agree the kill criterion before the first van is leased."
| Check the interviewer runs | Summary close | Answer close |
|---|---|---|
| Sentence one answers "roll out to all 40?" | No; it announces a summary | "No, not to all 40. The 14." |
| One number carries the decision | Nine numbers, none attached to a decision | $33,000 against minus $9,000 |
| One risk that could flip the call | "Risks to look into" | Two-store density read; 12 orders a day is the tripwire |
| Next step a manager could start Monday | None | Rank by density, two waves, kill criterion agreed up front |
I would score the first close a 2 of 5 and the second a 4 or 5. Same analysis; different question answered.
When the close gets interrupted
What the interviewer says after your first sentence tells you what they are testing.
| What you hear | What it tests | The 15-second answer |
|---|---|---|
| "What would change your mind?" | Whether you know your weakest assumption | "Orders per day. Below 10, even an urban store does not cover its van." |
| "The CEO wants all 40 anyway." | Whether you hold a position without pretending the data moved | "Then sequence it: the 14 first, a suburban wave only if orders hold above 12. The arithmetic does not change because the CEO wants it to." |
| "One sentence." | Whether sentence one was the answer | Repeat sentence one. If you cannot, you did not have one. |
| "Good, thanks." | Nothing | Stop. Do not add a sixth risk. |
The second row separates offers. Changing your recommendation because an interviewer raised an eyebrow is scored worse than being wrong; changing it because they gave you a new number is scored as judgment.
Practice this today: the echo drill
Take three cases you have already done. For each, write the prompt sentence at the top of an index card and, underneath, your first sentence using the same verb and one number. Record a full close into your phone: decision, two reasons, risk, next step. Count the words. Over 160, cut reason two. Under 100, you skipped the risk or the next step.
Then change the question and keep the facts. Re-close Northwind as if the CEO had asked "how do we double wash-and-fold revenue in two years." The facts do not move; the first sentence changes completely.
CoachNed's synthesis drill gives you a finished case's facts and scores only the close. The live voice case with Ned runs about eighteen minutes and ends with a seven-score debrief in which synthesis is its own line. The five-minute first rep works without an account. Everything is open for seven days, no card; then $120 for a recruiting season or $49 a month.
CoachNed is independent and is not affiliated with McKinsey, BCG, Bain, or any other firm named in this article.
Frequently asked questions
How long should the final recommendation in a case interview be?
About 60 seconds, or 130 to 150 words at a deliberate pace. Given thirty seconds, deliver sentence one and the one number; the rest follows if they let you continue.
Should I give a recommendation if I did not finish the analysis?
Yes, and put what is missing in the risk beat: "I would not sign the lease at this rent; the piece I have not seen is weekend footfall, and above 40% of weekday I would reopen the question." Declining to decide because the data is incomplete is scored as declining to decide.
What is the difference between a summary and a synthesis in a case interview?
A summary reports what you looked at, in the order you looked at it. A synthesis answers the client's question using only the facts that bear on it. Anyone who watched the case can give a summary; only someone who decided can give a synthesis.
Sources
- BCG Interview Preparation Brochure (November 2018, PDF) — "synthesize your thoughts" as a tested ability; "go back to the original question." Checked 2026-09-24.
- FashionCo. case, Bain & Company — Bain's good answer (one sentence, one number) and better answer (risks and next steps). Checked 2026-09-24.
- Preparing for the Case Interview, Bain & Company — "Clarify the objective: make sure you understand the question before diving in." Checked 2026-09-24.
- GlobaPharm practice case, McKinsey & Company — the case framed as one question, "Should GlobaPharm acquire BioFuture?" Checked 2026-09-24.
- Barbara Minto alumni profile, McKinsey & Company — Minto at McKinsey 1963–1973, the internal manual, the 1985 book. Checked 2026-09-24.
- Army Regulation 25-50, Preparing and Managing Correspondence (PDF) — paragraph 1-38: "single rapid reading" and "bottom line up front." Checked 2026-09-24.
- Yuan, Liberman, and Cieri, speaking rate in conversation, Interspeech 2006 (PDF) — Switchboard corpus at about 196 words per minute overall, 164 turn by turn. Checked 2026-09-24.