Interviewer-led vs candidate-led cases: what actually changes

Which format am I actually going to get?

Almost every question about case interview format is really a question about control. Who decides what happens next? In an interviewer-led case, the interviewer does. They arrive with a sequence of questions already written down, ask them in order, and hand you what you need when they decide you need it. In a candidate-led case, you do. You are given a situation and a question, and the next stretch of the interview is yours to organise.

The pattern most candidates have heard is broadly right: McKinsey leans interviewer-led, while BCG and Bain lean candidate-led. That is a real difference in house style, it is discussed openly, and it is worth preparing for. It is also softer than the prep forums suggest.

Real rounds mix the two. A candidate-led case will still contain moments where the interviewer takes the wheel and asks a pointed question with a single right answer. An interviewer-led case will still contain a question broad enough that you have to build a structure unaided. Practice varies by office, by round, and most of all by the individual interviewer, who is a person with habits rather than a script-reading machine. Experienced-hire rounds, final rounds run by partners, and cases built around a live client problem all drift towards the middle.

Read the last row of that table twice, because it is the point of this guide. The two risks are not variations of one theme. They are opposites. Which means preparation that only ever rehearses one format actively trains you into the failure mode of the other.

How does an interviewer-led case actually run?

An interviewer-led case opens like any other: a short prompt, a client, a question. The difference shows up about ninety seconds later, when the interviewer asks something specific and you realise it was question one of six.

The sequence is fixed before you arrive. It commonly moves through a structuring question, one or two quantitative questions built on exhibits, a question that asks for ideas rather than arithmetic, and a closing recommendation. The interviewer already knows which exhibit follows which question, and will produce it without being asked. Transitions are brisk and often unannounced: you will finish an answer and hear "good, let's move on", with no invitation to comment on where the case should go next.

What good looks like

Answer the question asked, completely, then stop. That sentence is most of the craft. Completely means the number plus the one line of interpretation it needs, not the number alone. Then stop means you resist the urge to bolt on the two adjacent thoughts you also had, because the interviewer has a question three that covers one of them and a question five that covers the other.

The second habit is signposting inside a single answer rather than across the case. You cannot say "and then I would move to costs", because that is not your call. You can say "there are three things I would want in this structure, and the third is the one I would start with", which shows the same judgement inside the boundary you have been given.

The specific trap

The trap is trying to run the case anyway. It comes from good instincts trained in the wrong room. A candidate who has drilled candidate-led cases for months will answer question one and then say "so the natural next step is to look at the customer segments, could I see data on that?" The interviewer, who has question two ready and it is not about segments, now has to steer you back. That correction costs you twice: it eats clock, and it reads as not listening.

The second version of the trap is quieter and more expensive: answering a broader question than the one posed. Asked for contribution per store, you deliver contribution per store, a view on the bakery's strategic role, and a comment on the brand. Two of those three were not requested. In an interviewer-led case the questions have been sequenced deliberately, so a broader answer is not generosity. It is noise laid over a question that had a specific target.

How does a candidate-led case actually run?

A candidate-led case hands you the pen. After the prompt, the sequence is yours: clarify what you have been asked, restate the objective in a measurable form, take a moment to build a structure, say where you want to start and why, ask for the data that branch needs, work it, draw the implication, then decide whether the next branch is still worth the time. When you have enough, you close without being asked to.

The interviewer's role changes accordingly. They are holding the answer key and releasing it on request. Ask for something that exists and you get it, often as an exhibit that was sitting in the folder all along. Ask for something that does not exist and you will be told to reason without it, which is itself part of the assessment. Long stretches can pass with the interviewer saying almost nothing, and that silence is not disapproval.

What good looks like

Visible ownership without thrashing. Ownership means the interviewer never has to ask what you are doing, because you said it before you did it: "I want to check whether this is a volume problem or a price problem, so could I see revenue split by both?" Without thrashing means you finish what you start. A branch opened, worked and closed with a stated implication is worth more than three branches half-explored because each new number suggested a new idea.

The other marker of a strong candidate-led case is that the structure survives contact with the data and is visibly amended when it should not. Saying "my third branch was going to be the cost base, but given what this exhibit shows about mix, I am dropping it and going to pricing instead" is a stronger move than either sticking to the plan out of loyalty or silently abandoning it.

The specific trap

The first trap is waiting to be led. A candidate trained only on interviewer-led cases lays out a structure, then pauses expectantly. Nothing comes. They pause again. Eventually the interviewer prompts, and in doing so takes back a decision that was being scored as yours. Repeat that four or five times and the transcript reads as a passenger's, however good the individual answers were.

The second trap is the mirror image: wandering because nobody stopped you. The freedom that makes the format satisfying also removes the guard rails. Nobody tells you that you have spent eleven minutes on market sizing when the question was about a make-or-buy decision. You are meant to notice. The absence of interruption is not a signal that the path is right, and candidates read it as one all the time.

What does the rubric reward differently in each format?

This is where the guide earns its keep, because the honest answer is that the rubric does not change at all. The scorecard we mark against has three dimensions and fourteen skills, each marked 1 to 5, where 3 means you met the first-round standard. Every one of those fourteen cells exists in both formats. What changes is how much of your evidence lands in each cell, and therefore which weaknesses are easy to hide and which are impossible to.

Interviewer-led loads precision, listening and rigor

Precision and listening sit in the presence and communication dimension, and an interviewer-led case is close to a purpose-built test of both. Precision is scored on whether you answered the question actually asked, in the units it was asked in, without padding. In a format where every turn is a direct question, you generate a fresh precision judgement every ninety seconds. Twenty of those in half an hour, and there is nowhere for a habit of over-answering to hide.

Listening gets the same treatment. When the interviewer controls transitions, they also control the nudges: a repeated word, an emphasis on "per store", a deliberately loose summary of what you just said to see whether you correct it. Miss those and the cell goes unfilled, because you never had the option of setting your own direction and being right by luck.

Rigor rises too, for a mechanical reason. The maths is put in front of you rather than requested, so you cannot route around a quantitative weakness by choosing branches that happen to be qualitative. Whatever the exhibit contains, you are doing it, out loud, at the interviewer's pace.

Candidate-led loads framing, creativity and synthesis

Framing and creativity are both analytical sub-skills, and both get far more evidence in a candidate-led room. Framing is scored on whether the structure fits this problem, breaks into parts that do not overlap, and names a place to start. In an interviewer-led case you produce one structure on request and are then steered. In a candidate-led case your structure is the thing that actually governs the next twenty minutes, so the interviewer sees not just whether it was well built but whether it was any use.

Creativity behaves similarly. Asked directly for ideas, most prepared candidates can produce a decent list. Choosing, unprompted, to open a branch that nobody suggested, and justifying why it matters commercially, is a different and better piece of evidence for the same cell.

Synthesis is the one to watch, because in a candidate-led case nobody prompts your close. In an interviewer-led case the final question is usually an explicit invitation: the client is on the phone, what do you say. In a candidate-led case you have to decide that you now have enough, stop analysing, and deliver a recommendation first, with reasons attached to numbers, a named risk and a next step. Candidates who have only ever been invited to synthesise frequently run the clock out mid-analysis, and an unfinished case cannot be scored on a close that never happened.

What does each format punish?

The clearest way to see this is to take one moment and play it two ways in each format. Same candidate, same knowledge, different instinct about who is steering.

Interviewer-led: the exhibit is handed to you

The interviewer slides an exhibit across and says: "So, what do you make of this?" That question has a target, and it is narrower than it sounds.

The weak version loses precision for padding, loses listening for trying to redirect, and offers no number where the exhibit was plainly put there to produce one. The strong version reads the exhibit, states what it means, commits to an interpretation, and stops. It also does not ask to see anything, because in this format that request costs more than it can possibly return.

Candidate-led: you decide you need the exhibit

Now the same content, other way round. You have finished a branch, you have an implication, and nothing has been handed to you.

The weak version hands the agenda back, which in a candidate-led room means handing back the marks. It also asks a question the interviewer cannot usefully answer without doing your job for you. The strong version names the reasoning, makes a specific request, and pre-commits to a route if the data does not exist. Notice that it stays strong whether or not the exhibit turns out to be available, which is the tell of a well-formed request.

One symmetry worth naming: neither format punishes asking for a number to be repeated, and neither punishes taking a moment to think. Those are guard rails in the rubric, not concessions. What gets punished is the unrequested detour in an interviewer-led room, and the request for direction in a candidate-led one.

How do I prepare for both without splitting my preparation?

The good news is that the split is much smaller than it looks. Structuring, case maths, commercial judgement, reading an exhibit, sense-checking a number, delivering a recommendation in sixty seconds: all of that transfers whole. You are not preparing twice. You are preparing once and then rehearsing two specific joints.

Those joints are the opening ninety seconds and the handling of transitions. Everything else you already practise serves both formats.

If you are practising with us, the roster runs both formats on purpose, so you can alternate without changing anything about how you prepare. Yuki Brennan-Mori and Marin Lefèvre-Asch run interviewer-led cases: they will pose their own question sequence, hand exhibits over unprompted, and move you on. Mira Vasquez-Holt, Lior Sarkis, Adrian Caldwell-Pierce and Soren Hallgrimsson run candidate-led cases, where the silence after the prompt is yours to fill.

You never have to guess which one you are walking into. The pre-flight screen before every case states the format and what it means for you, so the diagnosis in step one is free and you can spend the session on the skill instead. All of our coaches are AI personas, built and voiced to the specification of ex-MBB interviewers, and we say so everywhere rather than letting anyone assume otherwise. The first case is free, and the fastest way to find out which format exposes your weaker half is to run one of each.