In a typical first round of eight candidates, I push back on all eight at least once, and about half the time I agree with the number I am attacking. That is not a trick. It is the cheapest measurement an interviewer can take, and it measures one thing: how far you move per unit of new information.
Most guides treat "I disagree" as a verdict to survive. It is a calibration test with two symmetrical failure modes. The fold moves 100% on zero new information. The wall moves 0% on real new information. The sheet rewards neither. It rewards proportion: a move exactly as large as the objection justifies, with the size of the move said out loud.
By the end of this piece you should be able to classify an objection by its information content in about three seconds, size your response with a number rather than a mood, and say what would change your mind before anyone asks.
What the challenge is for, in the firms' own words
Firms publish little about this moment, but what they do publish points one way. McKinsey lists "uphold the obligations to engage and dissent" among its values, and a 2023 McKinsey article on dissent says that "people shouldn't agree if they don't agree". Bain's case-interview page tells candidates to "listen and adapt" to interviewer cues, and its university hiring team wrote in March 2026 that it wants people who "take a position supported by clear reasoning", adding that interviewers "are not here to trick or mislead you." BCG's preparation page says there "isn't always a single 'right' answer" and that the quality of your reasoning matters most.
A firm that asks its own people to dissent upward wants to know, before it hires you, whether you can hold a position against someone senior and drop it when the facts change. The partner across the table runs that test in miniature, sometimes deliberately and sometimes because they genuinely disagree. You cannot tell which, and the correct response is identical either way.
No firm publishes a scoring sheet for this moment; anyone who tells you what a specific firm's sheet says is guessing. The generic sheets I have used reward four things: did you verify before moving, was the move proportional, did you state the new position with a reason, and do you know what would change your mind. Everything below is built around them.
Sort the objection by how much information it carries
Before you respond, work out what you were actually given. Most objections carry far less information than their tone suggests.
| What the interviewer said | Information content | Right-sized move | What it sounds like |
|---|---|---|---|
| "Hmm. Are you sure about that?" | Zero. Tone only. | Verify the line, then hold. | "Let me check. 18,000 units at $45 is $810,000. I'll hold that." |
| "That feels high to me." | An opinion, no data. | Ask for their number or offer a range; show the sensitivity. | "What would you use? From 8% to 15% the answer holds. Above 20% it flips." |
| "The client's actual churn is 30%, not 15%." | New data. | Recompute in the open; say whether the recommendation changed. | "At 30% the pool halves to $1.2 million and build no longer clears the hurdle; I'd partner instead." |
| "Your two buckets overlap." | A logic flaw. | Repair the structure. Do not defend it. | "Agreed, list price sits in both. I'll move it under commercial." |
| "The CEO doesn't care about margin. She cares about the brand." | A change of objective. | Re-anchor the recommendation, not the analysis. | "Then the question is what the trial does to brand, not to year-one contribution." |
The first two rows contain no new facts. A candidate who changes a number on them has updated on nothing, and I mark it that way. The last three contain real information; a candidate who does not move on them is not showing conviction. They have stopped listening.
The tell for a fold is one word: "actually," followed by a new number, with no calculation in between. When I hear that sequence I know the number came from my face and not from the page.
The forty seconds after "I disagree"
Pushback feels like it demands an instant answer. It demands a visible process, which takes about forty seconds.
| Seconds | What you do | What I see from the other chair |
|---|---|---|
| 0 to 3 | Stop talking. One breath. | You did not flinch. |
| 3 to 8 | Name the object: the number, the assumption, the logic, or the objective. | You heard which thing I challenged. |
| 8 to 25 | Do the work where I can see it: re-run the line, or compute how far the answer moves under the alternative. | Method. This is the part the sheet is built for. |
| 25 to 35 | State the move in one of three words: hold, move to, or flip. Give the reason. | Proportion. |
| 35 to 40 | Give the kill criterion: what would have to be true for you to change your mind. | You know your own uncertainty, which is rarer than being right. |
Silence while your pencil is moving is fine for ten seconds, sometimes fifteen. Silence while your eyes are on me is not fine for three. And the question "what would you use?" is legal, useful, and almost never asked. If the objection carried no data, asking for the data is the only honest way to update.
Worked example: a price rise that survives the challenge
Harbor Lane Bakeries is an invented chain of 60 shops. It sells 2.0 million sourdough loaves a year at $8.00. Ingredients, packaging, and per-loaf labor come to $6.00, so each loaf contributes $2.00 and the line earns $4.0 million of contribution a year. Should Harbor Lane raise the loaf to $8.40, a 5% increase?
You assume a 5% volume loss. Volume falls to 1.9 million loaves, contribution per loaf rises to $2.40, and new contribution is 1.9 million times $2.40, or $4.56 million: an uplift of $560,000. You recommend the increase.
The interviewer: "A 5% volume loss feels optimistic. I think you'd lose far more than that."
Row two: an opinion, no data. Three responses, in the order I most often hear them.
The fold: "You're right, that's probably too aggressive. Maybe the increase doesn't work." A recommendation abandoned because I frowned.
The wall: "Five percent is a standard assumption for a staple product." Possibly true, entirely unsupported, and silent on whether the recommendation is fragile.
The proportional answer computes the one number that settles it: how much volume can Harbor Lane lose before the rise stops paying? It pays as long as new volume times $2.40 is at least the old $4.0 million, so volume must stay above 1.67 million loaves. Harbor Lane can lose a sixth of its volume, 16.7%, before the rise is a wash. The shortcut: breakeven volume loss equals one minus old unit contribution over new unit contribution, here one minus 2.00 over 2.40.
| Volume loss | Loaves sold | Contribution at $2.40 | Change versus today |
|---|---|---|---|
| 0% | 2,000,000 | $4,800,000 | plus $800,000 |
| 5% | 1,900,000 | $4,560,000 | plus $560,000 |
| 10% | 1,800,000 | $4,320,000 | plus $320,000 |
| 12% | 1,760,000 | $4,224,000 | plus $224,000 |
| 16.7% | 1,666,667 | $4,000,000 | zero |
| 20% | 1,600,000 | $3,840,000 | minus $160,000 |
Said aloud: "You may be right that 5% is light. The recommendation doesn't depend on it. It holds for any volume loss under about 17%, and at 10% it still nets $320,000. Above a sixth of volume, I flip. Do we know what happened the last time they moved price?"
The interviewer gives you something: "In 2023 they raised the loaf 5% and volume fell 12%."
Row three: real data. Recompute in the open. 2.0 million times 0.88 is 1.76 million loaves, times $2.40 is $4.224 million, an uplift of $224,000. The recommendation survives, but 12 against a breakeven of 16.7 is thin. The proportional move is neither "hold" nor "flip." It is "move to": "I'd still raise the price, but stage it through ten shops first, because a repeat of 2023 leaves under five points of headroom."
Count the moves. The opinion changed how you expressed your confidence and nothing else. The datum moved the recommendation one notch. That is what proportional looks like.
Scoring the three answers
My approximation of a generic sheet, scored zero to two on each of the four things above. It is not any firm's rubric.
| What the sheet rewards | The fold | The wall | The proportional answer |
|---|---|---|---|
| Verified before moving | 0 | 1 | 2 |
| Move proportional to the information given | 0 | 0 | 2 |
| New position stated with a reason | 1 | 1 | 2 |
| Knows what would change their mind | 0 | 0 | 2 |
| Total out of 8 | 1 | 2 | 8 |
The column that surprises candidates is the second. People who stonewall believe they are at least demonstrating conviction. From my chair the fold and the wall are the same failure with different body language: neither shows me the candidate thinking about the objection. The proportional answer is not clever. It is merely complete.
Why smart people fold, and why the room is worse than the lab
In 1955 Solomon Asch published the experiment everyone half-remembers: 123 American college students judging which of three lines matched a standard, each seated among confederates coached to give the same wrong answer. Alone, people erred under 1% of the time. Against a unanimous wrong majority, they went along on 36.8% of the critical trials; about three quarters yielded at least once, and every one who yielded underestimated how often. Against a single opponent, though, subjects erred only 3.6% of the time. One dissenting peer barely moved anyone.
Your interviewer is a single opponent, so on Asch's numbers you should be nearly immune. You are not, because the lab used peers and no stakes, and the room uses a partner and an offer. My estimate from the chair, and only an estimate: on a zero-information challenge, between a third and a half of first-round candidates change their number, closer to Asch's unanimous-majority figure than his single-opponent one. Like his subjects, they do not remember doing it. Treat the 1955 figures as an order of magnitude from one population, not as constants.
Asch's yielders explained themselves with "I am wrong, they are right." The interview version is "they must know something I don't." Sometimes the interviewer does, which is exactly why you ask what it is before you move.
One more finding is usable. A single supporting partner cut conformity to a quarter of its unanimous-majority level. You get no partner in the room, so your page has to be it: a number written down with its calculation beside it is harder to abandon than one held in your head. Write the arithmetic; do not do it silently.
Outside consulting, Amazon's leadership principles file the same skill under "Have Backbone; Disagree and Commit": leaders "do not compromise for the sake of social cohesion," then "commit wholly" once a decision is made. McKinsey's dissent article supplies the wording for a wall that happens to be right: "I still believe in my alternate solution, but I'm grateful for the opportunity to contribute to this process, and I respect that you have the final say." Hold the content. Give away the decision. Candidates who conflate the two either fold or sulk.
Ned's rule. Move exactly as far as the new information moves the answer, and say the distance out loud. A full reversal on "are you sure?" and a zero move on a real number are the same mistake from my side of the table: neither one shows me thinking.
Practice this today: rehearse the objection, not the case
Take any case you have already solved and write down the three numbers the recommendation depends on. For each, compute the breakeven: how far it has to move before the recommendation flips. Then say the sentence aloud, three times each: "I'd hold unless volume loss goes above 17%." Fifteen minutes, and you have rehearsed the 8-to-35-second window on a case you already understand.
Then run it live. The voice case with Ned on /interview runs about eighteen minutes and ends with a seven-score debrief; whenever one of your numbers gets questioned, run the forty-second sequence out loud and listen for the word "actually." New to CoachNed? /start gives you three typed turns and instant scores in five minutes, no account needed. The recompute step is a speed problem, which is what /drills/math is for; the one-line "hold, move, or flip" statement is what /drills/synthesis is for.
Everything is open for seven days, no card; then $120 for a recruiting season or $49 a month. CoachNed is independent and not affiliated with McKinsey, BCG, Bain, Amazon, or any other firm named here.
Frequently asked questions
Should I change my answer when the interviewer pushes back?
Only by as much as the pushback justifies. Fix an arithmetic error immediately, move within a range if the interviewer offers a different assumption, and flip the recommendation only when new data or a new objective carries it past the breakeven you computed. A challenge with no information in it earns a verification and a restatement, not a new answer.
What if the interviewer is wrong?
Say so with the calculation rather than with adjectives, and grant them the decision without granting the content: "I've re-run it and get the same $810,000, so I'll hold that; tell me what you'd change." Interviewers challenge correct numbers on purpose, so being challenged is not evidence you were wrong.
Is pushback a sign the case interview is going badly?
No. I challenge every candidate at least once, and I challenge answers I agree with as often as ones I doubt. A challenge tells you which part of your answer is being scored next; it says nothing about the score so far.
Sources
- Our purpose, mission, and values — McKinsey & Company — "uphold the obligations to engage and dissent" is a stated firm value. Checked 2026-09-24.
- Into all problem-solving, a little dissent must fall — McKinsey, February 15, 2023 — "people shouldn't agree if they don't agree" and the withholding-assent wording. Checked 2026-09-24.
- Preparing for the Case Interview — Bain & Company — "Listen and adapt" and "Think aloud." Checked 2026-09-24.
- How to stand out in Bain's consulting recruiting process — Bain careers blog, March 11, 2026 — "take a position supported by clear reasoning"; interviewers "are not here to trick or mislead you." Checked 2026-09-24.
- Case Interview Preparation — BCG Careers — "there isn't always a single 'right' answer." Checked 2026-09-24.
- Opinions and Social Pressure — Solomon E. Asch, Scientific American, November 1955 (PDF) — 123 subjects; under 1% error alone; 36.8% under unanimous pressure; 3.6% against a single opponent; a supporting partner cut conformity to a quarter; yielders underestimated their own conformity. Checked 2026-09-24.
- Leadership Principles — Amazon.jobs — "Have Backbone; Disagree and Commit." Checked 2026-09-24.