I have scored several hundred cases and asked, by my count, about seven questions. The client changes, the chart changes, the accent of the partner next to me changes. The questions do not. And every one of the seven is scored for something other than its literal answer: "what do you see in this chart?" is not a chart-reading test, and "what would you tell the CEO?" is not a request for a summary.
That is the whole thesis. Candidates prepare for hundreds of questions because question banks are long, then lose points on the seven because they answer the words instead of the line on the sheet the words belong to.
By the end you will be able to hear any prompt, name the line it is scoring, and open with the sentence that earns the mark.
Five lines on the sheet, seven ways to fill them
Firms do not publish their scoring sheets, and I will not describe any one firm's. Every sheet I have used, over more years than I will admit to, had four to six lines, always some version of the same five: structure, numeracy, judgment, synthesis, and presence. Each line gets a mark, usually one to five. Hire or no-hire is a conversation about the pattern, not a sum.
You can see the five lines in what the firms do publish. McKinsey says its problem-solving interview exists to understand "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). Count the verbs. That is a scoring sheet with the numbers removed. BCG's page says it in fewer words: "what matters most is how you approach the problem and the quality of your reasoning" (BCG, Case interview preparation).
The seven questions are the fastest way to put a mark on each line.
| What you hear | The line it scores | The answer that misses | The answer that lands |
|---|---|---|---|
| "How would you approach this?" | Structure | A profitability tree recited from memory | The decision in one sentence, three or four branches specific to it, and which you would open first |
| "What would you want to know first?" | Structure (problem definition) | "More information about the market" | One or two questions whose answers would change the work: the metric, the horizon, the constraint. Then silence |
| "What do you see?" (an exhibit lands) | Judgment, via analysis | A reading of the chart, top left to bottom right | The one comparison that matters, what it does to your hypothesis, and what you would check next |
| "What is the number?" | Numeracy | Silent arithmetic and a bare number | Setup out loud, assumptions named, a number with a unit and a sanity check, then what it means |
| "What else could be driving this?" | Judgment (breadth and practicality) | Nine items in the order they occurred to you | Two or three buckets, depth on the one that matters, and a word on what is actually doable |
| "Are you sure?" | Presence, then judgment | An immediate reversal | Separate arithmetic, assumption, and conclusion; re-check the one challenged; hold or update, with a reason |
| "So what do you tell the CEO?" | Synthesis | A chronological recap | The answer, two reasons, one risk, one next step. A decision, not a summary |
Only one of the seven is scored on what it literally asks for, and even numeracy is a hygiene line: a right number earns the mark; a right number with no meaning attached leaves the judgment line blank.
The answer key Bain published, and what it gives away
The clearest public evidence is Bain's own FashionCo sample case. For the brainstorming step: "A good answer includes different types of ways to increase revenue. A better answer includes the pros, cons and practicality of each idea." For the recommendation, the good answer picks the option "projected to generate $50M more," and the better answer "would include additional risks and next steps" (Bain, FashionCo case).
Read that from the interviewer's chair. The brainstorm was scored on judgment, not on the count of ideas. The recommendation was scored on whether you behaved like someone who has to live with it on Monday. Neither better answer needed more knowledge than the good one. Both needed the line answered rather than the words.
Who asks the seven varies by firm. The sheet does not
Candidates sort firms into "interviewer-led" and "candidate-led." Useful vocabulary, and candidate vocabulary: none of the firm pages linked here uses either phrase. The pages do show a difference in who says the questions out loud.
In a McKinsey-style discussion, by consistent candidate report, the interviewer asks the seven as a sequence of prompts; McKinsey's own page promises only "a fast-moving discussion." Answer the prompt in front of you, and keep a private structure so the close still works. In a Bain or BCG-style discussion, mostly you ask them: Bain's sample case tells you to "propose where you would like to start" once you have walked through your structure. If nobody asks "what do you see?", say it anyway. If nobody asks for a recommendation, give one at minute 25. In BCG's Casey online case, a chatbot asks; BCG describes an interactive case guided by Casey that takes about 30 minutes and "simulates the Case Interview experience" (BCG careers FAQ).
Where you lead, you must ask yourself the seven, in order, unprompted. The most common way I see a strong candidate fail a Bain-style case is not a wrong answer. It is reaching minute 28 having never asked question seven.
The prompts that do not sound like questions
Bain's own tips page says "Listen and adapt: pay attention to interviewer cues and adjust accordingly" (Bain, Case interview preparation). Here are the cues, from my side of the desk: what those moments usually meant when I said them, not what any firm's manual says.
| What you hear | What it usually meant when I said it | What to do |
|---|---|---|
| "Interesting." | Nothing. I was writing | Keep going |
| Silence after your number | I am waiting for the so-what. The numeracy mark is in; the judgment mark is not | Say what the number means for the decision, in one sentence |
| "Anything else?" | There is one more item on my list and I am checking whether you have it | Give one more, the most material one, then stop. Do not empty the bucket |
| "Take a minute." | An offer, not a trap. I prefer 45 seconds of silence and a structure to 45 seconds of talk and none | Take it. Say "I will take a minute," then take one, not four |
| "Let's move on." | Either the line is filled or I have given up on it; for you, both mean the same | Move. Do not circle back; the repair costs more than the mark |
| "Why?" after your recommendation | I want the causal chain, not more data | One sentence of mechanism: X drives Y, and Y is what the client is paid on |
Ned's rule. The first sentence of every answer should be scoreable on its own. If I cannot put a mark on the sheet after your first sentence, the rest of your answer is you negotiating for one.
One question, two answers, same arithmetic
The thesis, with numbers you can reproduce on a napkin. The company is invented.
Redwood Self Storage runs 60 facilities in the US Southeast, 600 units each, at an average rent of $150 per unit per month and 90% occupancy. The board wants an 8% rent increase across the portfolio. Last year a 5% increase cost two points of occupancy, from 92% to 90%. The interviewer says: "Assume tenants respond the same way, scaled. What happens to monthly revenue, and what do you tell the CEO?"
The arithmetic, which both answers below get right:
| Step | Calculation | Result |
|---|---|---|
| Units in the portfolio | 60 x 600 | 36,000 |
| Revenue today | 36,000 x 90% x $150 | $4.86M per month |
| New rent | $150 x 1.08 | $162 |
| Expected occupancy loss | 2 points per 5%, so 3.2 points per 8% | 86.8% |
| Revenue after the increase | 36,000 x 86.8% x $162 | $5.06M per month |
| Change | $5.06M less $4.86M | About $200k per month, $2.4M a year, a 4% lift |
| Break-even occupancy | $4.86M divided by (36,000 x $162) | 83.3%, a fall of 6.7 points |
Answer A: "Revenue goes from about 4.86 million to about 5.06 million, so up around 200,000 a month. So the increase works and I would recommend it."
Answer B: "Raise it. You gain about $200k a month, call it $2.4M a year, a 4% lift. The more useful number is the break-even: occupancy would have to fall below 83%, nearly seven points against the two you lost last year, before the increase costs you money. Two caveats: that two-point history is a single data point, and a tenant who leaves costs move-in incentives to replace. So stage it: the 20 fullest facilities first, watch move-outs for 60 days, then roll it out."
Scored on the five lines, one to five each:
| Line | Answer A | Answer B | Why |
|---|---|---|---|
| Structure | 2 | 3 | Neither set up the calculation aloud; B at least ordered its answer |
| Numeracy | 4 | 4 | Same arithmetic, same result |
| Judgment | 2 | 5 | B computed the number the CEO actually needs: how wrong can we be before this hurts |
| Synthesis | 2 | 5 | A summarized; B decided, and gave an action a client could start on Monday |
| Presence | 1 | 4 | "So it works" is a shrug. B sounds like someone who will be in the room when it goes wrong |
| Total | 11 | 21 |
Answer B is about 60 words longer and contains no arithmetic that A lacks. The whole gap is which question each candidate thought they were answering. Only one of them was on the sheet.
Where candidates lose the mark they had already earned
Two patterns account for most of the leakage I have seen, and neither is a knowledge gap.
Answering the literal question. The chart gets described, the number gets stated, the list gets listed. Each is a complete answer to the words and an incomplete answer to the line. The fix is mechanical: after any number, add "which means"; after any observation, "so I would"; after any list, "and I would start with."
Treating pushback as a verdict. "Are you sure?" is the cheapest question I own. It costs me nothing and tells me more than any other prompt about how you will behave with a client who disagrees. In a typical first round of eight, I would watch two or three candidates reverse a correct position within five seconds. That is a mark lost on judgment and another on presence, for an answer that was right.
Structure is the only line where more preparation reliably helps. The other four are habits of answering, and habits are built in reps, not in reading.
Practice this today
One assignment, 30 minutes. Take any case, live or recorded, and keep a sheet beside you with the five lines. Each time the interviewer speaks, write the line the prompt is scoring before you answer. Afterwards, replay only your first sentences and score each on its line, alone, as if the rest of the answer had never happened.
The fastest way to get a reference score is the five-minute first rep: three typed turns of a case, each scored, no account needed. When you want the full thing, the live voice case runs about 18 minutes with a seven-score debrief, which is close enough to a real sheet to be uncomfortable. If the leaks are on the judgment and synthesis lines, the synthesis drill works the answer-first reflex in isolation.
Everything on CoachNed is open for seven days without a card; after that it is $120 for a recruiting season or $49 a month. CoachNed is independent and has no affiliation with McKinsey, BCG, Bain, or any other firm named here.
Frequently asked questions
How many questions are asked in a case interview?
Fewer than you think. A 25 to 30 minute case usually holds five to eight interviewer prompts, drawn in my experience from the same seven types: approach, clarification, exhibit, number, brainstorm, pushback, and recommendation.
Do McKinsey, BCG, and Bain ask different case interview questions?
They ask the same seven in different ways. McKinsey's format tends to put them to you as discrete prompts, Bain's and BCG's discussions expect you to raise most of them yourself, and BCG's Casey online case has a chatbot ask them.
Should I memorize answers to a list of case interview questions?
Bain's own tips page says "Listen and adapt: pay attention to interviewer cues and adjust accordingly" (Bain, Case interview preparation). Here are the cues, from my side of the desk: what those moments usually meant when I said them, not what any firm's manual says.
No. Memorize the five lines and the first sentence that earns each one. A framework recited from memory at "how would you approach this?" is the most recognizable failure pattern from the interviewer's side, and Bain's own page says it wants how you "build on ideas as the conversation evolves," which a script cannot do.?
The clearest public evidence is Bain's own FashionCo sample case. For the brainstorming step: "A good answer includes different types of ways to increase revenue. A better answer includes the pros, cons and practicality of each idea." For the recommendation, the good answer picks the option "projected to generate $50M more," and the better answer "would include additional risks and next steps" (Bain, FashionCo case).
Are brain teasers still case interview questions?
Rarely at the large strategy firms, in my experience, and neither Bain's nor BCG's preparation page mentions them. If one appears, treat it as a numeracy prompt: set it up aloud, state assumptions, produce a number with a unit. Do not prepare 40 of them.
Sources
- McKinsey & Company, Interviewing — what the problem-solving interview assesses. Checked 2026-09-24.
- BCG Careers, Case interview preparation — approach and reasoning matter most. Checked 2026-09-24.
- BCG Careers, Mexico office FAQ — the Casey-guided online case, about 30 minutes. Checked 2026-09-24.
- Bain & Company, Case interview preparation — the "Listen and adapt" tip. Checked 2026-09-24.
- Bain & Company, FashionCo sample case — the "good answer" and "better answer" keys. Checked 2026-09-24.