What to take away
- Almost none of the twelve costliest case interview mistakes is an ability problem: they are habits, and habits move in days
- Every mistake maps to one named skill on the scorecard, so the fix is a specific rehearsal rather than more general practice
- Silent maths and a hedged close cost more than a wrong number, because an interviewer can only score what they can hear
- The most expensive judgement error is diagnosing a decline inside the company without first asking what the market did
- Four things candidates fear are not deductions: repeating a number, pausing to think, being decisive and wrong, and solving it your own way
Where do structures go wrong before the analysis even starts?
Almost none of the twelve mistakes below is an intelligence problem. They are habits: things candidates do because nobody told them to stop, or because a prep guide told them to start. That distinction matters more than it sounds, because habits move on a timescale of days. Across the transcripts we mark, someone who stops doing maths silently, or who replaces a hedge with a recommendation, does not need another month of cases. They need three or four cases where they deliberately do the other thing until the other thing is automatic.
It also matters that each mistake has an address. Our rubric splits a case into three dimensions and fourteen skills, and every habit below lands on exactly one of them: framing, commercial judgement, rigor, creativity and synthesis on the analytical side, presence, precision, listening and relationship-building on the communication side. Knowing which cell a habit costs you is what turns vague self-criticism into a drill.
The first three live in the opening five minutes, which is the most heavily observed part of the case. An interviewer forms a working view of your structuring in that window and then spends the rest of the case testing it.
Reciting a memorised framework at a problem it does not fit
In the room it sounds recited: "I'd like to look at revenue, then costs, then the market, then competition, then the customer." The tell is that nothing in it names the client, and it would have been said whatever the prompt was. This costs framing, which marks whether the structure fits this problem rather than whether a structure exists at all. The fix is mechanical enough to rehearse: write the client's actual question at the top of your page, then require every branch to be a candidate answer to that question. If a branch cannot complete the sentence "this matters because it would tell us whether…", it is decoration. Cut it before you speak.
Naming no hypothesis and no starting branch
It sounds like "so those are my four areas", followed by silence while you wait to be told where to go. The content may be fine, but a structure with no entry point reads as unfinished, and it quietly hands the interviewer a decision that was yours to make. This costs framing, and often presence with it. The fix is one sentence, always in the same shape, appended to every structure you ever present: "My hunch is that this is a cost problem rather than a price problem, so I'd start with cost per unit, because that is where the biggest swing sits." Rehearse the sentence rather than the structure. The structure changes every case; the closing sentence does not.
Building a tree so generic it would fit any case
"Internal factors and external factors." "Supply and demand." "The company, the customer, the competition." These are not wrong, they are weightless: they would fit an airline, a hospital and a cement plant equally well, which is precisely the objection. This costs framing, and it quietly costs commercial judgement too, because a tree that fits everything reveals no view about what actually drives this business. The fix is the substitution test. Read your branches back to yourself and ask whether they would survive being applied to a completely unrelated industry. If they would, you are one level too high. Replace at least one branch with something true only of this client: their channel mix, their contract length, their single largest customer.
What makes case maths lose points even when the answer is right?
Case maths is not marked on arithmetic. It is marked on rigor, and rigor is assessed on an auditable chain: whether the interviewer could follow your working, whether your assumptions were said out loud, and whether the answer was checked against anything. That is why a candidate who arrives at the right number can still score a 2, and why a candidate who slips on a multiplication, notices, and corrects can score a 3 or a 4.
Doing the arithmetic silently and announcing only the result
Eight seconds of silence, a pen moving, and then "forty-three point two million". There is nothing for the interviewer to mark and nothing for them to rescue if the number is wrong. This costs rigor directly, because an unspoken chain is an unscoreable chain. The fix has a fixed order: say the equation, then the inputs, then the answer. "I'm going to take volume times price. Volume is 2.4 million units, price is 18 pounds, so revenue is 43.2 million." Rehearse it on trivial sums until narrating feels slower than thinking, because in the room it will feel slower and you need to do it anyway.
Never sense-checking a number against anything
It sounds like total confidence: the number lands, the candidate moves on, and nobody asks whether a market of that size is plausible for a country of that population. This costs rigor and commercial judgement together, since the second half of business instinct is knowing when an answer is too good to be true. The fix is one sentence after every number, and it should become reflex: compare the result to something already on the table. "That is about a fifth of the market we sized earlier, which feels high for a challenger, so let me check the volume input." A sense-check that finds nothing still scores. A missing one never does.
Losing the units, or drifting between millions and billions
The giveaway is a number with no noun attached: "so we get forty-three point two", or a quantity that starts as revenue in millions and reappears two minutes later as a billion. This costs rigor, and precision as well, because the interviewer now cannot use anything you said. The fix is written discipline: put the unit beside every number as you write it, never in your head, and say the unit out loud every single time you say the number. Add one habit in front of the calculation: name the order of magnitude you expect before you compute. "This should land in the tens of millions." You will catch a factor-of-a-thousand error in the moment rather than in the recommendation.
Which judgement mistakes do interviewers notice most?
Commercial judgement is the least drillable-looking skill on the scorecard and, in practice, the one that moves fastest once you know what it rewards. It is not general business knowledge. It is the ability to tell which branch matters, what a number implies for the client, and when something in the case does not add up. All three habits below are failures of attention rather than of knowledge.
Chasing the branch you are comfortable with rather than the one that matters
It sounds reasonable and it is the most common single misstep we see: "Let me start by sizing the market", offered to a client who knows their market perfectly well and has asked why margin fell four points. This costs commercial judgement, because branch selection is where judgement is visible. The fix is to make entry explicit and cheap to abandon. Before you go into any branch, say what you expect to find and what you will do if it is empty: "I'll spend three minutes on cost per unit, because a margin fall on flat volume points there. If cost per unit is flat, I'll drop it and move to price." Rehearsing that line also gives you permission to leave a branch, which is the harder half.
Reading an exhibit for its description instead of its implication
It sounds like narration: "This chart shows revenue by segment from 2021 to 2025, with four segments on the horizontal axis." The interviewer can see the chart. This costs commercial judgement, and synthesis too, because an exhibit is a question in disguise. The fix is a three-line habit you can rehearse against any chart you meet, including in a newspaper: name what moved, quantify it, then say what it means for the client's question. Take ten silent seconds to orient first, then speak. "Two of the four segments are flat and premium has halved. That is roughly 60 per cent of the total decline, so the problem is concentrated, not broad."
Ignoring whether the market moved when diagnosing a client's decline
It sounds decisive, which is what makes it expensive: "Volume is down 8 per cent, so our sales effort must be weak." A client losing share in a growing market and a client holding share in a shrinking one have almost nothing in common, and recommending the first fix for the second problem is a serious error of judgement. This costs commercial judgement. The fix is a single question asked before any internal diagnosis, every time, until it is automatic: "Before I look inside the company, do we know what the market did over the same period?" If nobody can tell you, say which answer would change your conclusion and carry both branches forward for a minute.
How do candidates lose points on communication alone?
A quarter of the score in a full interview sits in presence and communication, and unlike the analytical block it can be lost while your thinking is entirely sound. The two habits here are the ones that cost most, and both are audible in the first thirty seconds of a recording.
Answering the adjacent question, then padding when the answer is thin
Asked "what is the breakeven volume?", the candidate says "well, there are some risks around pricing that I'd want to flag first". Asked for two drivers, they produce a fourth and a fifth that are visibly weaker than the first two. Both cost precision, which marks whether you answered the question that was asked, in the form it was asked, without filler. The fix is a three-beat rehearsal: the answer, one sentence of why, then stop talking. Silence after a short answer feels like failure and reads as confidence. If you have two good reasons, give two. A third weak one drags the average of everything you just said.
Talking over the interviewer's nudge
It sounds like a monologue with an interruption in it: "and if we then look at the third branch…" while the interviewer is halfway through "interesting, though what about fixed costs?" Worse, interviewers sometimes plant a deliberately wrong summary of your own analysis to see whether you catch it, and a candidate in full flow agrees with it. This costs listening, and it is the cheapest cell on the whole scorecard to lose. The fix is a hard rule plus one rehearsed sentence. The rule: full stop the instant the interviewer speaks, mid-word if necessary. The sentence, used whenever they summarise: "Let me make sure I have that right. You said margin was flat, but my note says it fell two points, can we check which it is?"
What does a weak close cost, and what does a strong one sound like?
The close is the most valuable sixty seconds of the case and the most reliably wasted. It is where synthesis is scored, and synthesis has a defined shape: the recommendation first, then the reasons with numbers, then the risk, then the next step. Candidates lose it in three ways, and the three are usually the same candidate on the same day.
Closing with a hedge, a diagnosis, or an answer with no risk attached
The first face is the hedge: "there are several factors and I'd want to run more analysis". The second is the diagnosis with no action: a correct account of why margin fell, offered as though naming the cause were the same as advising the client. The third is an answer with no risk and no next step, which reads as overconfidence and gives a partner nothing to sponsor. All three cost synthesis. The fix is a four-beat close, rehearsed as a fixed shape so it survives being tired at minute twenty-eight: recommendation, two or three reasons with numbers in them, the single biggest risk to your view, the first thing you would do on Monday. Practise it out loud against a stopwatch until you land it inside a minute without notes.
“Hedged: There are several factors at play here. Cost per unit has moved and price is under pressure, so I think I'd want to run more analysis on both before I say anything too definitive.”
“Decisive: I would not enter this market. Three reasons. The market is 120 million pounds growing at 3 per cent, we would need roughly a fifth of it, around 24 million, to clear the hurdle rate, and the two incumbents already hold four fifths between them.”
“Decisive, continued: The biggest risk to that view is my volume assumption, which I built from channel data rather than primary demand. So the first step is two weeks of trade interviews to test it, and I would reopen the decision if achievable share came in above 20 per cent.”
The second version is not better because it is longer or more confident. It is better because it is falsifiable. It states a view, names the number that would overturn it, and gives the client something to do next week. Note also that the decisive close could be wrong on the facts and would still outscore the hedge, because a partner can work with a reasoned view and cannot work with no view at all.
| Mistake | Skill it costs | The fix to rehearse |
|---|---|---|
| 01 Reciting a memorised framework | Framing | Write the client's question at the top; every branch must be a candidate answer to it |
| 02 No hypothesis, no starting branch | Framing | Append one fixed sentence: my hunch is X, so I'd start with Y, because that is the biggest swing |
| 03 A tree that would fit any case | Framing | Substitution test: if a branch survives a different industry, replace it with something only true here |
| 04 Silent arithmetic | Rigor | Fixed order out loud: the equation, then the inputs, then the answer |
| 05 No sense-check | Rigor · commercial judgement | One sentence after every number, comparing it to something already on the table |
| 06 Lost units, or millions drifting to billions | Rigor · precision | Unit beside every number on paper and in speech; name the expected order of magnitude first |
| 07 Chasing the comfortable branch | Commercial judgement | Before entering a branch, say what you expect to find and what you'll do if it is empty |
| 08 Describing an exhibit, not reading it | Commercial judgement · synthesis | Three lines per chart: what moved, how much, what it means for the question |
| 09 Ignoring what the market did | Commercial judgement | Ask what the market did over the same period before any internal diagnosis |
| 10 Answering the adjacent question, then padding | Precision | Three beats: the answer, one sentence of why, then stop |
| 11 Talking over the nudge | Listening | Full stop the instant they speak; repeat any summary back before agreeing with it |
| 12 Hedging, diagnosing, or closing with no risk | Synthesis | Four-beat close inside a minute: recommendation, numbered reasons, biggest risk, Monday's first step |
Twelve is a long list to work on at once, and you should not try. Take the one that appeared most often in your last case, drill it for three cases until it is boring, and only then look at the next. A hundred short drills at /drills exist for exactly this: each one isolates a single skill so you are not rehearsing everything and improving nothing.
Which case interview mistakes are not actually mistakes?
Case-prep folklore has invented deductions that appear on no scorecard anywhere. This section is not a reassurance exercise. Across the transcripts we mark, the fear of these four does more measurable damage than the four things themselves ever could, because the candidate spends attention on avoiding a phantom penalty instead of on the case.
| What you have been told | What actually happens |
|---|---|
| Asking for a number again shows you weren't listening | Nothing is deducted. If it is noted at all, it is noted under listening as a positive |
| Long silences make you look slow | Framing judges the structure you commit to, not a half-built one. The pause is protected time |
| A wrong recommendation ends the case | A reasoned view that turns out wrong outscores a vague safe one on synthesis, every time |
| Not using a recognised framework looks unprepared | The quality of the thinking is scored, not the shape. A sound non-standard route can reach a 5 |
Asking for a number to be repeated
This is not a deduction and never was. Figures are said once, out loud, often over a slightly compressed video call, and using a number you are not certain of is a far worse move than asking for it again. A candidate who asks is doing what a competent colleague does. "Sorry, could you give me that revenue figure once more? I want to be sure I have it before I build on it" is a sentence you can use as often as you need to.
Taking a real pause to think before speaking
Sixty to ninety seconds of quiet to build a structure is normal, expected, and specifically accommodated by how framing is marked: the score reflects the structure you commit to, not a partially built one glimpsed halfway through. What does cost you is narrating your own process to fill the silence. Ask for the time, take it, then speak once. "Could I have a minute to lay out how I want to approach this?" is all it takes, and no interviewer has ever written it down as a concern.
Being decisive and turning out to be wrong
The scale here is worth being precise about, because it is where the folklore does the most harm. Scores run 1 to 5, where 3 means you met the first-round standard, and a 1 does not mean the skill appeared weakly. It means the skill never appeared at all. So a candidate who attempts the maths and fumbles it while talking through their assumptions scores a 2 or a 3. A candidate who says "I'm not confident with numbers, could you walk me through it" scores a 1 by never attempting it. The same logic governs the close: a committed answer that is wrong is scoreable, and a hedge is not.
Solving the case in a way no guide described
There is no expected shape you are being checked against. A 5 is explicitly reachable through a genuinely insightful non-standard move that achieves the same end as the conventional route, and interviewers enjoy those more than a well-executed template. What is scored is whether your route addressed the client's question, whether the reasoning held together, and whether you could explain it to someone who had not seen your page. If those three are true, the fact that your structure looks nothing like a textbook profitability tree is neutral at worst and a differentiator at best.
If you want to see which of the twelve you actually do, rather than which ones you suspect, run a case and read the per-skill marks. Our coaches are AI personas built by ex-MBB interviewers, and they mark against the same three dimensions and fourteen skills described here, with the evidence quote attached to every score. The first case is free, and one honest transcript will tell you more about your habits than another week of reading about them.
Common questions
What is the single most common case interview mistake?
Hedging at the close. Asked for a recommendation, the candidate offers factors and a request for more analysis instead of a view. It costs synthesis, which is one of the five analytical skills, and it is entirely fixable with a rehearsed four-beat close: recommendation, reasons with numbers, the biggest risk, the first next step. Silent maths runs a close second.
Why do candidates fail case interviews when they are clearly smart?
Because the scorecard marks visible behaviour, not private ability. Thinking done silently cannot be scored, a structure with no starting point reads as unfinished, and an answer to the adjacent question is marked as an answer to the wrong question. Scores also average, so one skill left at 1 or 2 can pull an otherwise strong case below the line while everything else looks fine.
What counts as a red flag in a case interview?
Four things get written down as real concerns: never committing to a recommendation, talking over the interviewer or missing a planted correction, stating an implausible number with no sense-check, and diagnosing a decline inside the company without ever asking what the market did. Note that being wrong is not on that list. Being unscoreable is what does the damage.
Is it a mistake to ask the interviewer to repeat a number?
No. This is one of the most persistent myths in case preparation and it is simply not a deduction on any scorecard we have used or built. Numbers are said once and often over a video call. Asking for a figure again is what a careful colleague does, and it is far better than silently building three minutes of analysis on a figure you misheard.
Does using a standard framework count as a case interview mistake?
Using one is fine. Reciting one at a problem it does not fit is the mistake. Framing marks whether your structure addresses this specific question, so a profitability tree is excellent on a profitability case and poor bolted onto a market-entry question. Borrow the shape, then cut the branches that do not apply and add at least one that is true only of this client.
How long does it take to fix these mistakes?
Most of them move within a week, because they are habits rather than knowledge gaps. Narrating your arithmetic, closing with a recommendation and stopping when a question is answered each need roughly three cases of deliberate practice to become automatic. Judgement habits, such as checking what the market did, take a little longer. Work on one at a time; drilling all twelve at once changes none of them.