In a typical first round of eight cases, two or three candidates lose points for the same reason in opposite formats: they wait for instructions when they should drive, or they keep driving when the interviewer has already chosen the next slide. The label "interviewer-led" versus "candidate-led" is a useful default map. It is not two different sports.
What the scoring sheet rewards barely changes. What changes is who picks the next branch. After this piece you will know the firm defaults, how to detect the live format in the first five minutes, and one habit that works in both: state the so-what and the next move before you go quiet.
The myth that costs offers
Prep folklore treats McKinsey as a quiz show and BCG or Bain as an open mic. Firms do not publish those labels. What they do publish is closer than candidates expect.
McKinsey's interviewing page describes a Problem-Solving Interview built around a client scenario: structure an ambiguous challenge, work with facts and data, form conclusions, and talk through a fast-moving discussion. Their public practice cases (Beautify, Talbot Trucks, GlobaPharm) are sequenced questions with exhibits and math prompts. That is the interviewer-led pattern in the wild: you still structure and synthesize, but the interviewer advances the agenda.
BCG's case prep page asks you to structure the approach, ask thoughtful questions, analyze data, do quick calculations, and identify what matters. Bain's case interview hub calls the case a structured conversation about a real business problem and says they are watching how you think, structure, and build on ideas as the talk evolves. Deloitte's case tips walk through clarify, assumptions, findings, recommendation, and next steps, and explicitly ask for a framework plus an initial hypothesis.
None of those pages say "passenger mode is fine" or "ramble until data appears." They all score ownership of the logic. The format only changes who queues the next exhibit.
What each format actually looks like in the room
| Signal in the room | Interviewer-led default | Candidate-led default |
|---|---|---|
| After your opening structure | Interviewer names the first area or hands you an exhibit | Silence, or "Where would you like to start?" |
| Data | Arrives with the prompt | You request it with a reason |
| Math | Embedded in a specific question | You choose the calculation that tests the hypothesis |
| Mid-case | New prompt closes the prior chapter | You synthesize and pick the next branch |
| Endgame | Final prompt for recommendation or risks | You close when the evidence supports a call |
Firm defaults I use when coaching (not firm policy documents): McKinsey leans interviewer-led; BCG and Bain lean candidate-led or hybrid; Big Four strategy arms are mixed by practice and office. Treat that as a prior. The person across the table can flip it.
The same sheet, two failure modes
When I score a case, I am listening for four things whether I am steering or not: structure that fits the objective, clean math, business judgment, and synthesis that a client could act on. Format does not rewrite that list.
| Failure | Interviewer-led version | Candidate-led version |
|---|---|---|
| No ownership | Answers each prompt, never says what it means | Collects facts without a path |
| Weak structure | Framework dumped once, then forgotten | Generic MECE tree never tied to a hypothesis |
| Math theater | Correct number, no implication | Correct number, no decision |
| Wandering | Fights the interviewer for control of the agenda | Asks for data out of curiosity |
Passenger mode and tourist mode share a root cause: the candidate never claims the next insight. In interviewer-led cases, claiming it sounds like a one-line so-what after each prompt. In candidate-led cases, claiming it sounds like "I would test X next because Y."
Ned's rule. Treat three seconds of silence as permission to lead. Treat every redirected prompt as a closed chapter: synthesize in one sentence, then follow. Never fight the interviewer for the steering wheel, and never sit quietly when they have handed it to you.
Worked example: HarborLine Fitness (same numbers, two formats)
Prompt: HarborLine operates 40 mid-market gyms. Membership revenue has been flat at $48M for two years. Operating profit fell from $6.0M to $3.6M. Why, and what should management do?
Shared opening (both formats). Restate the objective: diagnose the profit drop and recommend one priority move. Clarify that "flat revenue" means membership dues only, not retail or PT. Structure: profit = revenue − cost; with revenue flat, prioritize cost and mix (retention quality, discounting buried in revenue, labor, rent, marketing).
Interviewer-led path. Interviewer: "Break down the cost side." You: labor, rent, marketing, equipment depreciation, and other. Interviewer hands a table: labor rose from $18.0M to $20.4M; rent flat at $9.6M; marketing up $0.6M; other flat. Math you can reproduce:
| Line | Year −2 | Year 0 | Delta |
|---|---|---|---|
| Revenue | $48.0M | $48.0M | $0 |
| Labor | $18.0M | $20.4M | +$2.4M |
| Other opex (implied) | $24.0M | $24.0M | $0 |
| Profit | $6.0M | $3.6M | −$2.4M |
Profit drop equals the labor delta exactly. So-what: "The entire $2.4M profit decline is labor. I would next check whether headcount or wage rates drove it, then whether utilization fell." Interviewer may then give headcount: 1,200 FTE to 1,280 FTE at constant average wage. Implication: ~6.7% more staff on flat membership — a utilization problem, not a wage spike.
Candidate-led path. After structure, silence. You lead: "Given flat revenue, I will start with contribution margin and labor as a share of revenue. Can I get labor and membership count by year?" Same table arrives because you asked for it. Same math. Same so-what. Then you choose the next branch: "Before I recommend cutting staff, I want attrition by tenure — if we lost high-tenure coaches, wage mix could fool us." You own the agenda; the interviewer only supplies facts.
Same case. Same numbers. Different person choosing the next slide. Same scoring criteria.
How to detect the format in five minutes
- Open the same way every time: objective, clarifying questions that unlock the structure, then a short issue tree tied to a hypothesis.
- After you present the tree, pause.
- If the interviewer picks a branch or slides an exhibit across, shift to interviewer-led mode: answer the prompt fully, synthesize, stop.
- If they nod and wait, shift to candidate-led mode: name the first test, ask for the data that would falsify your hypothesis, run it.
- Re-check every time they speak. Hybrids are common. A "quiet" BCG interviewer can still redirect you mid-case.
Phrases that signal control without sounding robotic:
| Moment | Line that works |
|---|---|
| Answering a prompt | "Two parts: the calculation, then what it means for the recommendation." |
| Taking the lead | "I would next test utilization versus wage rates, because that decides whether we cut hours or renegotiate contracts." |
| Getting redirected | "Understood. Parking the broader cost tree; focusing on labor now." |
| Requesting data | "To falsify the utilization story, I need FTE and visits per member." |
Firm defaults without the folklore
| Firm (usual pattern) | What their public materials emphasize | Prep bias |
|---|---|---|
| McKinsey | Sequenced sample cases; fast-moving problem-solving discussion | Drill prompt-by-prompt answers and exhibit reads |
| BCG | Step-by-step structure, questions, data, calculations | Drill leading the issue tree under light guidance |
| Bain | Structured conversation; practice cases like Coffee Shop Co. and FashionCo. | Drill hypothesis-led data requests and recommendations |
| Deloitte | Five-step approach; framework and initial hypothesis | Drill short cases that still close with next steps |
Office, round, and interviewer style still override the table. Bain's own tips tell you to clarify, structure, think aloud, and adapt to cues — which is hybrid language, not a pure open mic.
Practice both without two personalities
You do not need a McKinsey persona and a Bain persona. You need one habit practiced under two constraints.
| Drill | Constraint | What you are training |
|---|---|---|
| Prompt stack | Partner feeds five unrelated questions in 12 minutes | Synthesize after each answer |
| Silent partner | Partner only answers data requests | Lead for 25 minutes without prompting |
| Redirect mid-tree | Partner steals the agenda twice | Close the open loop before following |
| So-what timer | After every chart, one sentence in under 10 seconds | Kill math theater |
Run a live case on CoachNed if you want the hybrid version: the voice interview on /interview will interrupt, redirect, and still expect you to own the next insight. Everything is open for seven days, no card; then $120 for a recruiting season or $49 a month.
Drill this tonight
Take HarborLine's numbers above. Run it twice in 40 minutes: once with a partner who only asks scripted prompts, once with a partner who stays silent unless you request data. Score yourself only on whether you stated a so-what and a next move after every analysis. If that is your weak spot, start with the free five-minute case on /start and a structure drill on /drills/structure.
Frequently asked questions
Is McKinsey always interviewer-led?
No. McKinsey's published sample cases are question-driven, and that is the usual pattern, but individual interviewers vary by office and round. Prepare to lead if they leave space after your structure.
Are BCG and Bain always candidate-led?
Usually more candidate-led or hybrid than McKinsey, based on how their careers pages describe the conversation. Plenty of BCG and Bain interviewers still hand you exhibits and steer. Watch the silence, not the logo.
Which format is harder?
Neither, if you synthesize. Interviewer-led is harder for people who go passive. Candidate-led is harder for people who collect facts without a hypothesis. Most candidates fail the format they practiced less, not the "harder" one.
Do Big Four interviews use the same split?
Deloitte and peers publish case or scenario interviews that ask for structure, hypothesis, and next steps; formats mix by practice, and some processes add group or written cases. Confirm with your recruiter and default to leading until redirected.
Should I memorize different frameworks for each format?
No. Build a fit-for-purpose structure once, then change only who chooses the next branch. Memorized framework swaps are how candidates sound scripted when the interviewer pivots.
Sources
- Interviewing at McKinsey — Problem-Solving Interview description and sample-case library (Beautify, Talbot Trucks, and others). Checked 2026-09-24.
- BCG case interview preparation — Official expectations: structure, questions, data, calculations, reasoning over a single right answer. Checked 2026-09-24.
- Bain case interview preparation — Case as structured conversation; clarify, structure, think aloud, adapt to cues; Coffee Shop Co. and FashionCo. practice cases. Checked 2026-09-24.
- Deloitte case and scenario interview tips — Five-step approach; framework and initial hypothesis; logic over a single right answer. Checked 2026-09-24.
CoachNed is not affiliated with McKinsey, BCG, Bain, Deloitte, or any other firm named above.