The person deciding whether you pass a group case is not listening to the group. She has two names on her sheet, yours and one other, and a grid with five or six rows. Each time you speak she makes a small decision: is this worth writing down, and which row does it go in.
That is the whole exercise. Silence leaves rows blank, and a blank row cannot be scored up later no matter how good your closing was. Talking over people fills the rows with the wrong evidence. The candidates who pass produce a modest number of attributable, classifiable moves, spread across the half hour, that a tired observer can file without effort.
This guide covers who still runs group exercises (fewer firms than the prep industry implies), how the scoring works behind the clipboard, how much airtime you actually have, six moves that reliably land in a row, and an exhibit trap you can rehearse with a friend tonight.
Who still puts candidates in a room together
Group cases live mostly in Big Four early-careers processes and assessment centers, not in strategy-firm case rounds. Here is what the firms' own pages say as of September 2026. All of it changes without notice, so your invitation outranks this table.
| Firm and process | What the firm's page says | Group element? |
|---|---|---|
| Deloitte UK, Final Stage Assessment | Two exercises: a 30 to 40 minute discussion of a prepared topic and a skills and motivation interview of about 50 minutes, in person for graduates (Deloitte UK) | Not listed |
| PwC UK, virtual assessment center | A "virtual immersive assessment centre" scored against core skills and attributes; exercises not itemized (PwC UK) | Not specified |
| KPMG UK, Launch Pad | In-person half day of assessed and non-assessed activities, dated 17 November 2026 to 5 May 2027 (KPMG UK); a graduate profile on the same site describes a group exercise, a written exercise, and an interview (KPMG profile) | Yes, per KPMG's own graduates |
| EY UK, Experience Day | In-person event with "a range of activities and exercises" before EY assessors (EY UK); EY Australia spells it out: a group activity with two to four other students on a fictional case with no single right answer (EY Australia) | Yes |
| Bain, consulting roles | A digital assessment, then case interviews (Bain) | Not on the page |
| Oliver Wyman | Conversational and case interviews (Oliver Wyman) | Not on the page |
| Roland Berger | Interviews plus case studies (Roland Berger), with recruiting run country by country (Roland Berger Q&A) | Reported by candidates in some offices; not confirmed by the firm |
Two things stand out. The Deloitte group exercise that older candidate accounts describe is not on Deloitte UK's current page. And the firms that do use groups describe them the way EY does: no single right answer, and the interest is in how the group works. That is not modesty. It is a description of the scoring sheet.
What the observer is doing with the pen
Assessment centers are a documented craft. The International Taskforce guidelines, published in the Journal of Management in 2015, define the method as multiple trained assessors who observe and record behavior during simulations, classify what they recorded against the constructs being measured, and only then rate it. Assessor training calls the sequence ORCE, observe, record, classify, evaluate, with recording done live and classifying done afterward on a form that lists behavioral indicators under each competency (South African assessment centre guidelines, 5th edition).
In the rooms I have sat in, that looks like this. Each observer gets two candidates, sometimes three. She writes close to verbatim while you talk, because she cannot classify and listen at once. Afterward she tags each line to a row. A row with no tagged lines is "not observed," and in every integration meeting I have attended, not observed is treated as the bottom band. Nobody argues you up on the basis of what you might have said.
The rows vary by firm, but the grids I have used converge on something like this:
| Row | Positive indicator she can tag | Negative indicator she can also tag |
|---|---|---|
| Structuring | Names the decision and the metric that settles it | Lists ideas with no decision to hang them on |
| Using data | Reads an exhibit and states what it changes | Repeats a number already on the table |
| Communicating | Two to four sentences, claim then support | Turns over a minute; restating own point |
| Working with others | Builds on a named person's point; asks a quiet person a specific question | Interrupts; takes the marker unasked |
| Driving to an outcome | Calls the time; forces a recommendation | Opens new threads in the last five minutes |
Notice what is missing: "got the right answer." The group's answer is recorded once, for the group. Your score is the tags next to your name. Evidence has to be attributable and classifiable: a subject (you), a verb the observer recognizes, and ideally a number or a name.
Your airtime is about 800 words
Run the arithmetic on the common format: 30 minutes, five candidates, four exhibits. Most designs give three to five minutes of silent reading, leaving about 25 minutes of talk. Divided evenly, that is five minutes each. At a conversational 140 words a minute, parity is about 700 words. Aim a little above it, six minutes, and your entire scored performance is about 840 words, spread over ten to twelve turns of 30 to 45 seconds. A short memo, delivered in pieces.
| Share of 25 talking minutes | Minutes | Words at 140 a minute | How it reads on the sheet |
|---|---|---|---|
| 10 percent | 2.5 | 350 | Rows left blank; "quiet" in the margin |
| 20 percent (parity) | 5 | 700 | Enough for every row if the moves are chosen |
| 25 to 30 percent | 6 to 7.5 | 840 to 1,050 | Active without dominating |
| 40 percent and up | 10 or more | 1,400 or more | Dominance noted; the collaboration row suffers |
Those readings are my estimate from the observer's chair, not a firm's rule. The research on the underlying bias is less ambiguous. A 2020 study in The Leadership Quarterly filmed 33 leaderless student groups of four to ten people, 256 participants in all, and timed every utterance in their ten-minute planning sessions. Speaking time correlated 0.67 with how many "emerged as a leader" nominations a person received from peers, the effect survived controls for measured intelligence and personality, and the authors put it at about 39 seconds of speech per additional nomination. Twenty-six participants never spoke at all. Those raters were fellow participants, not trained assessors, and a grid exists precisely to dampen the effect. But your observer is still a human with a page, and a page with nothing on it does not get dampened into a pass.
The same paper points to the ceiling: it cites Woolley and colleagues' 2010 collective-intelligence work, which found that groups with more uneven speaking turns performed worse across tasks. Observers know this from experience. The group that let one person run it produces a thinner answer, and the person who ran it gets the blame in the write-up. So the target is a band, not a maximum: your share, slightly above, spread across the session, never the most in the room by a wide margin.
Six moves that get written down
Each of these is short, attributable, and lands in a known row. You do not need all six; four, at the right moments, is a strong sheet.
| Move | When | What it sounds like | Row it fills |
|---|---|---|---|
| Frame the decision and the metric | Minutes 0 to 3 | "The board wants a close list. The test for each club is whether it covers its own direct costs, not the reported profit line." | Structuring |
| Propose the split | Minutes 3 to 5 | "Four exhibits, five of us. Dan and I take club economics, Priya takes membership, you two take the leases. Back together in six minutes." | Driving to an outcome |
| Bring one number that changes something | Minutes 8 to 15 | "Footnote two says head office is allocated equally and fixed. That flips two of the three closures. Here is the arithmetic." | Using data |
| Pull in the quiet person with a specific ask | After minute 8 | "Priya, you had the membership trend. Does Marsh Lane's decline support a fix rather than a close?" | Working with others |
| Challenge once, with data, then move | When the group chases the wrong metric | "We are optimizing reported profit and the board cares about cash. Quayside is the only club with negative contribution. Can we test that before we lock the list?" | Structuring, Working with others |
| Call the clock and force the rec | Minutes 20 to 23 | "Seven minutes left. Proposal: close Quayside, review Marsh Lane and Eastgate, flag head-office cost as the real problem. Objections?" | Driving to an outcome |
The phrasing matters less than the shape: a claim, a number or a name, an invitation. "We should probably look at costs" has no subject to attach to you and gets no tag. "I think we need to be structured here" is all subject and no content and gets tagged in the wrong row.
Ned's rule. If the observer could not write your last contribution down in one line and file it under a competency, it did not happen. Speak in lines that file themselves: a claim, a number, a name.
A worked exhibit trap you can rehearse
An original pack of the kind group exercises favor. Harborline Fitness runs twelve gyms in one region, and the board wants a recommendation on which of its three loss-making clubs to close. Exhibit two shows the four clubs under discussion, in millions of dollars a year. Footnote two says head-office costs of 4.8 million are allocated equally, 0.4 per club, and are fixed for the medium term.
| Club | Revenue | Direct costs | Contribution | Allocated head office | Reported profit |
|---|---|---|---|---|---|
| Marsh Lane | 1.60 | 1.30 | 0.30 | 0.40 | -0.10 |
| Quayside | 1.10 | 1.15 | -0.05 | 0.40 | -0.45 |
| Eastgate | 2.00 | 1.75 | 0.25 | 0.40 | -0.15 |
| Northfield | 2.40 | 1.70 | 0.70 | 0.40 | 0.30 |
The instinctive group answer arrives inside eight minutes: close Marsh Lane, Quayside, and Eastgate, "saving" 0.10 plus 0.45 plus 0.15, or 0.70 million. Someone writes 0.7 on the board and the room relaxes.
The footnote flips it. Closing a club removes its revenue and its direct costs, so the company keeps or loses the contribution line, not the reported profit line. Marsh Lane contributes 0.30, Eastgate 0.25, Quayside minus 0.05. Close all three and Harborline loses 0.50 million a year of contribution, while the 1.20 million of head-office cost that sat on those clubs lands on the remaining nine and makes two more look loss-making next year. The swing between the group's headline and reality is 1.20 million, exactly the allocated overhead of the closed clubs. That identity holds in every allocated-cost exhibit you will ever see.
The better recommendation: close Quayside only, put Marsh Lane and Eastgate under a fix-or-exit review with a contribution target, and tell the board that a 4.8 million head office on twelve clubs is the actual problem. In the room it takes about 40 seconds:
"Before we lock the three, one check. Footnote two says head office is allocated equally and fixed, so closing a club only saves its direct costs. Marsh Lane and Eastgate cover theirs, by 0.3 and 0.25. Only Quayside does not. Closing all three costs us about half a million of contribution a year and reallocates 1.2 million of overhead onto the other nine. I'd propose Quayside only and a review for the other two. Priya, does the membership trend support a fix at Marsh Lane?"
Count the tags: a data catch (Using data), a restated decision rule (Structuring), a proposal (Driving to an outcome), a named question to a colleague (Working with others). Four rows, 40 seconds, and the group's answer got better. That is the exchange rate to aim for. The tone that scores is "one check," not "you are all wrong."
Scoring the same 30 minutes for two candidates
Here is how the sheet tends to come out for two common profiles in the Harborline room. Ratings are illustrative, on the 1 to 4 scale most grids use, and they are mine, not any firm's.
| Row | Candidate A: 11 minutes, 15 turns | Candidate B: 5.5 minutes, 7 turns |
|---|---|---|
| Structuring | Two-minute framework speech on revenue, cost, and "strategic fit." Score 3 | Named the test at minute four: covers direct costs or not. Score 3 |
| Using data | Summed the three reported losses to 0.7; missed the footnote. Score 2 | Caught the footnote; showed the 1.2 swing. Score 4 |
| Communicating | Turns averaged 45 seconds, three over a minute; restated own rec three times. Score 2 | Turns of 20 to 40 seconds, each a claim with support. Score 4 |
| Working with others | Took the marker unasked; talked over Raj twice; asked no questions. Score 1 | Asked Priya for the trend; built on Raj's lease point by name. Score 4 |
| Driving to an outcome | Called the time and delivered the closing. Score 3 | Called the time at minute 20; offered a one-line synthesis and let A present it. Score 3 |
| Total | 11 of 20 | 18 of 20 |
Candidate A was not stupid, and the group's final answer was right because B fixed it. A still scores 11 because his collaboration row is a 1 with two negative tags beside it, and a 1 in that row is very hard to argue against in the integration meeting. A third profile is worth naming: the candidate who says nothing for fifteen minutes and then delivers a beautiful closing. That sheet reads 1, 1, 3, 1, 3. Nine. The closing was real; so were the blank rows.
Virtual rooms change the mechanics
PwC UK runs its assessment center virtually and EY moves its Experience Day online for summer interns, so a fair share of group exercises now happen on a video call. The scoring does not change; three mechanics do. Attribution gets harder for the observer, so make it easy: open turns with a name ("Building on Raj's lease point") and close them by handing to a name. Unmute latency punishes the quick interjection, so replace the interrupt with a raised hand and a four-word signal in chat ("one check on footnote two"), then speak when called. And take the shared document early: on a call, the person typing the recommendation is visibly attributable, and narrating while you type is the one form of scribing that costs no speaking turns.
Rehearse the moves this week
You cannot practice the social load of a group room alone, and CoachNed does not pretend otherwise; its cases are one interviewer and one candidate. What you can build alone is what makes the six moves possible: a private structure and a fast read of an exhibit, so that when your turn comes you have a number rather than a feeling.
The assignment: run one live voice case at /interview and read your structure and synthesis scores in the debrief. Then do ten exhibit reads at /drills/charts, each time saying aloud in one sentence what the chart changes about the recommendation. Finish with three synthesis reps at /drills/synthesis, timed to 30 seconds. That is the "bring one number" move and the "call the clock" move, drilled in isolation.
Then borrow humans. Give three classmates the Harborline table, assign one to observe with the five-row grid above and two names, and run 20 minutes. Rotate. Have the person arguing for three closures push back, and deliver the correction in under 45 seconds without sounding pleased with yourself. The observer's notes will teach you more about your habits than any guide, including this one.
Everything on CoachNed is open for seven days, no card; after that it is $120 for a recruiting season or $49 a month. CoachNed is independent and not affiliated with Deloitte, PwC, KPMG, EY, Bain, Oliver Wyman, Roland Berger, or any other firm named here.
Frequently asked questions
Do McKinsey, BCG, and Bain use group case interviews?
Not in their published processes. Bain's hiring page describes a digital assessment followed by case interviews, and neither McKinsey's nor BCG's public interview pages list a group exercise. Group cases are a Big Four and assessment-center format; treat any claim that MBB runs them as folklore unless your invitation says so.
How much should I talk in a group case interview?
Slightly more than your even share of the talking time, and never the most in the room by a wide margin. With five candidates and 25 minutes of discussion, that is about six minutes in ten to twelve short turns spread across the session, not one long speech.
Should I volunteer to be the scribe or timekeeper in a group exercise?
Take the timekeeper role if it is offered, because calling the clock is a scorable behavior. Take the marker only if you keep speaking; a silent scribe fills no rows. On a video call, typing the shared recommendation while narrating it is the one form of scribing that helps.
What if the group reaches the wrong answer?
Challenge once, with a number, then move with the group. Observers score your challenge and how you handled the disagreement, not whether you won it. Prosecuting the point for five minutes converts a positive tag in the data row into a negative one in the collaboration row.
Sources
- Early Careers Assessment | Deloitte UK — Final Stage Assessment as two exercises, a topic discussion and a skills and motivation interview, with no group exercise listed. Checked 2026-09-24.
- Assessment and selection process | PwC UK — a virtual immersive assessment centre as the penultimate stage. Checked 2026-09-24.
- Application process | KPMG UK — Launch Pad as an in-person half day, with 2026 to 2027 event dates. Checked 2026-09-24.
- KPMG UK graduate profile, Technology Engineering — a graduate's account of Launch Pad including a group exercise. Checked 2026-09-24.
- How to apply | EY UK — Experience Day as an in-person event with exercises before EY assessors. Checked 2026-09-24.
- Student program application hub | EY Australia — group activity with two to four other students on a fictional case with no single correct answer. Checked 2026-09-24.
- Interviewing | Bain & Company — digital assessment followed by case interviews; no group case mentioned. Checked 2026-09-24.
- Interview preparation | Oliver Wyman — conversational and case interviews. Checked 2026-09-24.
- Selection process | Roland Berger — interviews and case studies as the published format. Checked 2026-09-24.
- Q&A with our team | Roland Berger — recruiting organized by country. Checked 2026-09-24.
- Testing the Babble Hypothesis: Speaking Time Predicts Leader Emergence in Small Groups (MacLaren et al., 2020), Binghamton University ORB — 33 groups, 256 participants, a 0.67 correlation between speaking time and leader nominations, about 39 seconds per additional nomination, 26 silent participants, and the Woolley et al. (2010) citation. Checked 2026-09-24.
- Guidelines and Ethical Considerations for Assessment Center Operations (Rupp et al., Journal of Management, 2015) — multiple trained assessors observe, record, classify, and rate behavior. Checked 2026-09-24.
- Best practice guidelines for the use of the assessment centre method in South Africa (5th edition) — the ORCE sequence and structured rating forms with behavioral indicators. Checked 2026-09-24.