A McKinsey case is interviewer-led. You open a structure and a hypothesis; they choose the next exhibit and the next question. You still have to think out loud, quantify, and close with an answer-first synthesis. Fail the case or the Personal Experience Interview and you fail the loop.
This page is the format and scoring map. It is not the 10-case practice bank and it is not the PEI question list.
Last full pass: 24 August 2026. CoachNed is independent practice material, not affiliated with McKinsey.
How the slot actually runs
A typical client-facing interview is about 50 minutes: PEI for 10 to 20 minutes on one story, then a problem-solving case of roughly 25 minutes, then your questions. Some offices reverse PEI and case. Either way, they are scored as separate components.
Interviewer-led does not mean passive. You still:
- Restate the objective and the constraint.
- Build a two- or three-branch tree that is specific to this client.
- State a hypothesis and which branch you would test first.
- After every number, volunteer the implication before they ask.
- Close in 60 to 90 seconds: answer, one number, one risk, one next step.
What you give up is choosing the path. What you do not give up is choosing to think.
Before live rounds you sit Solve. In 2026 the common invitation is two modules: Redrock Study around 35 minutes and Sea Wolf around 30 minutes. Confirm the invite. Solve is a screen, not a substitute for this case.
What they are scoring (and what they hear)
| Dimension | What a pass sounds like | What a fail sounds like |
|---|---|---|
| Structure | Two or three testable branches, named priority | A memorized profitability template on a public-sector prompt |
| Hypothesis | A directional claim you are willing to kill | "I would like to look at revenue and costs" |
| Exhibit | Describe, quantify, implicate | Reads the axis labels and waits |
| Math | Equation out loud, units attached, then the number | Right number, invisible logic — or a silent 90 seconds |
| Synthesis | Answer first, one number, one risk | Recap of the analysis, hedge, no recommendation |
| Drive | After each question you propose the so-what | You wait because they are "leading" |
The exhibit habit is the one candidates least rehearse. McKinsey hands charts one at a time. BCG often dumps three to five. If you practised only candidate-led cases, the McKinsey room feels like someone stole the steering wheel. The correction is simple: after every handoff, say what the number means for the hypothesis.
Run that motion on a scored guided case, then isolate the weak skill on drills.
Worked example: non-revenue water, not a grocery chain
Prompt. A municipal water utility serving 1.2 million people has seen billed volume fall 4 percent in two years while production at the plants is flat. The city council wants a recommendation in 90 days: is this leakage, theft, meter error, or demand destruction — and what should they fund?
Opening structure (interviewer-led, so keep it short). Three branches: (1) physical loss (pipe bursts, night-time leakage), (2) commercial loss (meter under-registration, unbilled connections), (3) genuine demand decline (efficiency, weather, industrial shutdowns). Hypothesis: production flat plus billed volume down is a loss story, not a demand story, so test commercial vs physical before you talk about a rate hike.
Interviewer question. "Here is district metered-area data. What do you take from it?"
Exhibit (the one they would hand you). Night-time flow in the oldest 30 percent of the network rose from 18 percent of daily production to 27 percent. Industrial billed volume is flat. Residential billed volume is down 6 percent. Meter replacement in that old zone has been frozen for five years.
The math. Production is unchanged, so the 4 percent billed-volume hole is not "people using less water at the plant." Night-time flow jumping 9 points in a third of the network is the physical-loss candidate. If that zone is 30 percent of production, a 9-point night-flow increase on that slice is about 0.30 × 0.09 = 2.7 points of system production disappearing as unbilled water. That is most of the 4-point billed gap. The residual ~1.3 points can sit in meter error in the same old zone. A rate increase does not fix either.
Synthesis. "Do not raise tariffs first. Most of the 4-point billed-volume loss is unbilled water in the oldest third of the network: night-time flow there explains about 2.7 points, and frozen meter replacement can explain the rest. I would fund a 90-day leak survey plus a meter swap in that zone, and I would tell council that a tariff hike on leaking pipes taxes households for water nobody billed. The risk is that a large industrial user is on a special contract outside the exhibit — I would confirm that before locking the story."
That is the McKinsey shape: a client-specific tree, one exhibit, arithmetic with units, a recommendation that names a number and a risk. For more of that shape, with ten fully worked interviewer-led cases, use the practice bank. Official McKinsey cases (Beautify, Diconsa, Electro-Light, GlobaPharm, National Education, Talbot Trucks) live on McKinsey's interviewing page.
The five question types inside one case
McKinsey cases are a sequence of discrete scored asks, not one continuous debate.
Structuring. "How would you think about this?" Two minutes of silence. MECE branches, specific to this client, with a stated first test.
Exhibit. "What do you take from this?" Describe, quantify, implicate. Technique: reading charts.
Brainstorm. "What could be driving this?" Grouped ideas beat a laundry list. Two named buckets of four ideas outscore nine orphans.
Quant. Say the equation before you compute it. That is also how they save you from a wrong path.
Synthesis. Sixty to ninety seconds. Graded separately. Method: case interview synthesis.
Failure modes that are specific to this room
- Going mute because they are steering. Interviewer-led is not a quiz show. After every exhibit, you still own the implication.
- Template trees. A profitability framework on a regulator, a hospital, or a city utility reads as you did not hear the prompt.
- Math in your head with no setup. A slipped number with a clear equation scores better than a magic correct number.
- Synthesis that recaps. If the first sentence is not the recommendation, you already burned the 90 seconds.
- Skipping PEI. The case does not rescue a hollow story. Prepare McKinsey-deep: PEI guide.
- Letting Solve eat the week. Redrock plus Sea Wolf is a real calendar cost. Do not discover that the night before cases.
A four-week plan that does not duplicate the case bank
Week 1. One official McKinsey case (start with Beautify for format). One structure drill a day. Draft eight PEI stories; do not polish them yet.
Week 2. Two interviewer-led reps from the practice bank or CoachNed cases. After each, write the failure in one line and map it to a drill: math, graphs, or synthesis.
Week 3. PEI probes on your weakest story. One full timed case. If Solve is on the calendar, one Redrock and one Sea Wolf transfer session — not leaked clones.
Week 4. Partner-round difficulty: harder exhibits, less coaching, tighter synthesis. Stop collecting frameworks. Rehearse the close.
McKinsey will vary by office and year. Confirm the live invitation. This page is transferable method, not tomorrow's prompt.