What to take away
- Most case preparation is genuinely solo work, and only two things really require another party: pushback and honest marking
- Reading worked cases builds recognition, not fluency, and recognition is what deserts you the moment you have to speak
- Recording yourself out loud and listening back is the highest-return solo habit, and almost nobody does it
- Write the evidence line before the number: if you cannot point at the moment in the recording, the score does not stand
- Score the five analytical skills 1 to 5 and fix exactly one of them at a time, not three
- An AI interviewer gives you volume and consistent marking, but it is not a person and cannot tell you how you will land in a real room
Why does solo case practice usually plateau?
Solo preparation has a worse reputation than it deserves. The great majority of the work that makes a candidate better happens with nobody else in the room: building structures, doing arithmetic out loud, reading exhibits, saying a recommendation in sixty seconds until it stops sounding like a hedge. What working alone cannot do is narrow, specific and fixable. The trouble is that almost nobody diagnoses it, so they plateau and blame the format.
The first failure is that reading a case is not doing a case. Working through a casebook with the answer visible builds recognition: you learn what a fitting structure looks like, and you feel a pleasant click when the exhibit resolves. Recognition is real progress, but it is not fluency. Fluency is producing the structure under time, out loud, with no model answer on the page. The gap between the two is invisible from the inside, which is why a candidate who has read forty cases can still freeze on the first one they have to speak.
The second is that nobody interrupts you. Alone, you finish every sentence, follow every branch you started, and never get asked "so what?" halfway through. A real interviewer cuts across you, changes the question, and hands you a number you were not expecting. Practising without any of that means you rehearse a monologue and then get tested on a conversation.
The third is the one nobody wants to hear: you grade yourself generously by default, and not because you are dishonest. You score the case you meant to run rather than the one you actually ran. You know what you were about to say next, so you quietly give yourself credit for it. You know why you paused, so the pause does not count against you. Every one of those adjustments is invisible and every one of them inflates the mark.
And you cannot hear your own hedging. "It probably depends on a few things" feels, from the inside, like appropriate care. From the other side of the table it reads as the absence of a view, which is the most expensive habit on the analytical half of the scorecard.
What actually transfers when you work alone, and what does not?
It helps to be blunt about what each activity buys you. The third column below is the one worth reading twice, because it is the column nobody puts in a study plan.
| Activity | What it genuinely builds | What it cannot build |
|---|---|---|
| Reading worked cases | Recognition: what a fitting structure looks like, what gets asked next, how exhibits usually resolve | Fluency. You never produce anything under time, so nothing becomes automatic |
| Timed structure reps | Speed and range: turning a prompt into three or four branches that fit the problem, inside ninety seconds | Judgement about which branch matters, because nobody makes you defend the choice |
| Maths drills | Rigor and calm: clean growth and percentage work, assumptions said out loud, a sense-check at the end | The feeling of doing arithmetic while somebody waits in silence |
| Exhibit reads | Speed at "what does this chart say", and the habit of naming the insight before the detail | Handling a follow-up that contradicts the read you just committed to |
| Recorded closes | Synthesis: recommendation first, reasons with numbers, risk, next step, inside sixty seconds | Whether the close survives "and what if the board disagrees?" |
| Mock interviews with a partner | Pushback, interruption, and the felt pressure of another person forming a judgement about you | Consistency. A peer's marking is only as calibrated as the peer, which is usually not very |
That first row is worth sitting with, because reading is the default activity for most candidates. It is comfortable, it is easy to do on a train, and it produces a strong sensation of progress. It builds recognition and not fluency, and recognition is exactly what deserts you when you have to open your mouth. A workable ratio is one hour of reading for every three hours of producing out loud.
The cheap fix for the reading trap
Cover the answer. Read only the prompt and the exhibits, run the case yourself out loud from start to finish with nothing in front of you, and only then read the model answer, treating it as marking material rather than as a lesson. The same casebook page does roughly four times the work in the same twenty minutes.
How do you mark your own case without flattering yourself?
Marking your own work is a skill, and it has exactly one enemy: you already know what you meant. The protocol below exists to strip that away, by forcing the evidence to arrive before the number.
Record the attempt
A voice memo on your phone is enough, or a screen recording if the case has exhibits. Run the whole thing out loud with no stopping and no restarts. A restart is you editing the evidence before anyone has looked at it.
Replay or transcribe it
Listen back before you form any opinion, or paste an automatic transcript into a document. The transcript is the thing you mark. Your memory of the attempt is not admissible, because it quietly includes the sentences you almost said.
Write the evidence line before the number
For each skill, write one line describing the actual moment: "at 4:10 I said the market looked attractive and never said why". Only then write a score. A number written first is a number you will spend the next minute justifying.
Score the five analytical skills 1 to 5
Framing, commercial judgement, rigor, creativity, synthesis, marked against the anchors: 3 means you met the first-round standard, 4 is clearly above it, 2 is a real attempt with a gap a coach would flag, and 1 means the skill never appeared at all.
Pick exactly one skill to fix next
Not three. The next two or three sessions target that single skill with a named behaviour attached, for example "every number gets a sense-check said out loud". Then re-mark and see whether the cell actually moved.
Two failure modes are worth watching for. The first is scoring generously across the board and reporting a tidy 3.4 average: if every cell comes out at 3, the marking is not discriminating between anything and you have learned nothing you can act on. The second is the mirror image, a mood-driven 2 everywhere after a session that felt bad. Both mean you scored the feeling instead of the transcript.
Keep the scorecards. Six of them in a folder, dated, is the only honest progress measure available to somebody preparing alone, and it will tell you within a fortnight whether the skill you chose is moving or whether you have been drilling the ones you were already good at.
What is the single highest-return solo habit?
If you take one thing from this article, take this: run every attempt out loud, and listen back to it. Not read your notes back. Listen. It is the closest thing solo preparation has to an outside observer, and very few candidates do it, because it is unpleasant in a specific and unavoidable way. The first playback is genuinely uncomfortable. By the third it is just information.
It works because speech carries most of the signal an interviewer scores, and almost none of that signal survives in your notes. A structure written on paper looks decisive. The same structure spoken aloud can be full of rising intonation, half-abandoned branches and "sort of". The scorecard marks the spoken version, so the spoken version is the thing to inspect.
What to listen for
- Hedging words: probably, roughly, sort of, I think, it depends. Count them. A dozen in five minutes reads as no view at all, however good the underlying thinking was
- The gap between saying a number and explaining it. If a figure arrives with no audible chain behind it, rigor is unscoreable even when the answer happens to be right
- Whether the recommendation actually came first, or arrived third, after two pieces of context you thought were setting it up
- Filler: um, so, basically, and the long "and then" that buys thinking time in the middle of a sentence instead of before it
- How long you took before speaking. Silence before you begin is normal and expected; narrating your own thought process while you build is the part that costs you
Time the silences deliberately, because playback corrects the estimate in both directions. Thirty to ninety seconds to build a structure is fine, and the recording tells you whether you took that time cleanly or filled it with noise. It also tends to reveal the opposite: the pause that felt like an eternity was four seconds long, and the interviewer would not have noticed it.
Two practical notes on mechanics. Do not stop the recording to fix a bad sentence, because the recovery is the skill being tested and cutting it out removes the only footage of you doing it. And do not delete the weak attempts. Being able to hear the version of yourself from three weeks ago is worth more than any amount of self-assessment done from memory.
Where do you genuinely need another party?
Here is where working alone runs out. Four things need a second party, and no amount of self-discipline substitutes for any of them.
- Pushback under pressure: being asked "why?" three times about the same branch, and having to hold or revise a view while somebody watches you decide
- Being interrupted mid-sentence: interviewers cut in deliberately, and the skill is stopping, absorbing and redirecting without losing the thread you were on
- A wrong summary planted to see whether you catch it: interviewers sometimes play your reasoning back slightly incorrectly, and noticing it politely is a scored moment you cannot stage for yourself
- The felt experience of a real person judging you: the adrenaline is part of what is being tested, and it does not show up when the only listener is your own phone
There are three ways to get some of that, and they are not equivalent. Each has something the others cannot offer.
| Option | Cost and availability | Realism and feedback consistency | What it cannot do |
|---|---|---|---|
| Peer mocks | Free, but you need a matched partner and two diaries that agree. Two or three a week is a realistic ceiling | Realism is high, because a real person is genuinely judging you. Consistency is the weak point: a peer who has never interviewed anybody marks by taste | Tell you where the bar is. Two candidates can agree a case went well and both be comfortably below the standard |
| Paid human coaching | The expensive option, usually priced per session, and the good ones are booked out. Scheduling is the real constraint, not money | The highest ceiling on both counts. An ex-interviewer notices what a peer and a machine both miss, and can say how you will read to a real partner | Give you volume. At that price most candidates buy two or three sessions, which is not enough repetitions to change a habit |
| AI mock interviews | Cheap to free, available at two in the morning, and effectively unlimited repetitions | Decent on the mechanics, and consistent on the marking because the same rubric is applied every time. Consistency is its genuine advantage over a peer | Be a person. It will not notice everything a partner would, and it cannot tell you how you will come across to the human who has to decide about you |
That last cell deserves more room than a table gives it. An AI interviewer is a simulator. It applies a rubric consistently, it is not tired by the eighth case of the evening, and it will hold you to a sense-check every single time. It also misses things: the small social read, the moment your energy drops, the way one particular phrase would land badly with one particular partner in one particular office. It cannot tell you whether a real person would want you on their team, because it is not a real person and has no team. Volume and calibration, not a verdict.
The sensible sequence, if you can get all three, is roughly this: solo work for volume, a simulator for consistent marking and repetitions, peer mocks for the human pressure and the interruptions, and one or two sessions with an ex-interviewer late in the process, when a calibrated outside read is worth paying real money for.
For completeness, since this is our site: MBB Ready is an AI mock interviewer. The coaches are AI personas, designed and voiced by the two of us as ex-MBB interviewers, and we say so on every page, because the distinction in the paragraph above is real and we would rather you knew it up front than worked it out later. There is a free three-minute spoken taster at /try with no signup, the first full case is free, and a 30-minute case-only session is the default. Everything is marked on the same three-dimension, fourteen-skill rubric with 1-to-5 anchors, where 3 means you met the first-round standard. What we cannot do is tell you how you will land in a room with a real partner, and neither can anybody else selling you software.
What does a one-week solo plan look like?
Here is a week you can actually run. It assumes a casebook or a set of practice prompts, a phone that records, and no partner available. The shape matters: it front-loads production over reading, and it ends in a full attempt you mark yourself, so the week produces a scorecard rather than a feeling.
| Day | The hour | What you end up with |
|---|---|---|
| Day 1 | 20 min reading one worked case with the answer covered, then comparing · 25 min on three timed structures, ninety seconds each, spoken aloud · 15 min listening back to one of them | A baseline recording and one written structure you would defend |
| Day 2 | 30 min of maths drills, growth rates, break-even and percentage change, all said out loud · 20 min on two market sizings with every assumption stated · 10 min listening back | A short list of the arithmetic that slowed you down |
| Day 3 | 25 min on four exhibit reads, one sentence of insight before any detail · 20 min of structure reps on the same prompts · 15 min writing evidence lines for one of them | Four insight sentences you would say to an interviewer unedited |
| Day 4 | 40 min running half a case out loud, from the prompt through the first piece of analysis · 20 min marking framing and rigor with the evidence-first protocol | Two scored cells with quotes sitting behind them |
| Day 5 | 20 min on closes only, six of them, sixty seconds each, recommendation first · 25 min listening back and counting hedges · 15 min rewriting the two worst | A hedge count and one fixed opening line |
| Day 6 | 30 min drilling the one skill your marking says is weakest, narrowly · 30 min of rest, or behavioural stories if a full interview is coming | One named behaviour to carry into tomorrow's case |
| Day 7 | 30 min on a full case attempt, recorded end to end, no stopping · 30 min marking all five analytical skills against the anchors, evidence line before every number | A self-marked scorecard and exactly one skill for next week |
The week is designed to repeat. Week two is the same shape with day 6 swapped for whatever the day 7 scorecard flagged, and with harder prompts. Four cycles of that will move your marks further than four weeks of casebooks, because each cycle closes a named gap rather than adding general familiarity.
One caution about the plan, which applies to every plan of this kind. It builds the analytical half of the scorecard and the communication habits that show up in a recording. It does not build your behavioural answers, and it does not expose you to pressure. Book one session with a person at some point in the fortnight, even an awkward one with a stranger from a forum, and treat everything above as what makes that session worth their time.
Common questions
Can you really prepare for a case interview without a practice partner?
Mostly, yes. Structuring reps, maths, exhibit reads and closes all improve alone, and that is the bulk of the work. What a partner supplies is pushback, interruption and the pressure of being judged by a person. Get those from a peer, a coach or a simulator, and keep your volume of solo repetitions high, because volume is what turns a structure into a habit.
How many practice cases should I do before a first round?
There is no threshold that guarantees anything, and any honest number varies enormously with your starting point. A more useful measure is whether your self-marked scores average 3 or above on every analytical skill, with none of them stuck at 2. A candidate with fifteen carefully marked cases and a closed weakness is in better shape than one with forty unmarked ones.
Is reading casebooks a waste of time?
No, but it is heavily over-used. Reading builds recognition: what a fitting structure looks like, what tends to get asked next, how exhibits resolve. It does not build fluency, because you never produce anything under time. Cover the answer, run the case out loud yourself, then read the model answer as marking material. Roughly one hour of reading per three hours of producing is a sensible ratio.
How do I stop over-scoring my own practice cases?
Mark the transcript rather than the memory, and write the evidence line before the number. If you cannot point at the moment in the recording where a skill actually appeared, the score does not stand. The discipline sounds fussy and it is the whole trick: the number you write after quoting yourself is routinely a full point lower than the one you would have written first.
Do AI mock interviews work as case interview practice?
For volume and consistent marking, yes. A simulator applies the same rubric every time, is available whenever you are, and does not get bored on the eighth case. Its limits are real: it is not a person, it will not notice everything a human interviewer would, and it cannot tell you how you will come across to the partner deciding about you. Use it alongside human practice, not instead of it.
What should I do if I have nobody at all to practise with?
Record everything and mark it against fixed anchors, which is the part solo candidates skip. Typed drills marked against those same anchors cover the analytical work cheaply, and we publish a hundred of them at /drills. Beyond that, one session with a person before the real thing is worth more than five extra solo cases, so it is worth the awkwardness of asking a stranger on a forum.