The study that gave us the phrase "deliberate practice" asked elite violinists which of their activities mattered most for getting better. They did not say lessons, and they did not say performing. They said practice alone. Case candidates are told the opposite: a partner is the real thing, and solo work is what you do while waiting for one.
The reverse is closer to the truth. Nearly everything a case interviewer scores is a predictable mechanic that arrives on a schedule: an opening structure around minute two, an arithmetic step around minute ten, a recommendation in the last three minutes. Predictable mechanics are what solo repetition builds best. Solo practice fails for a different reason. Most of it is reading with a timer running, and reading satisfies neither of the two conditions that make practice count.
By the end of this piece you will know which scoring rows you can build alone, how to spend 48 hours so you get four to six times the opening reps, how to grade your own recording without flattering yourself, and the one skill for which you should still find a human.
Two conditions, and a casebook meets neither
Ericsson, Krampe, and Tesch-Römer set out the conditions in their 1993 paper. Learners must receive "immediate informative feedback and knowledge of results," and they must "repeatedly perform the same or similar tasks." Without adequate feedback, they wrote, "efficient learning is impossible and improvement only minimal even for highly motivated subjects." Their violinists, at every skill level, rated practice alone as the most relevant activity for improvement, and the two strongest groups logged almost three times as many hours of it as the group training to be music teachers. Practice alone, the authors noted, is the one high-value activity a learner can extend at will, because it needs no teacher and no audience.
Now hold a casebook against those conditions. Reading a case and its model answer is repetition of the wrong task, since the task on the day is speaking, with feedback on nothing, because you produced nothing to grade. It does feel like preparation. Roediger and Karpicke showed why in 2006: students who reread a passage four times were more confident they would remember it than students who read it once and tested themselves three times. A week later the rereaders recalled 40 percent and the self-testers 61 percent, after reading the passage 3.4 times to the rereaders' 14.2. Rereading buys confidence. Testing buys retention.
So the missing ingredient in solo practice is not a partner. It is a task you actually perform, out loud, followed by a score you did not award yourself by feel.
Ned's rule. If nobody heard it and nothing scored it, it was reading, not practice. Reading teaches you what a good answer looks like. It earns nothing on the sheet.
What the sheet rewards, and how much of it you can build alone
BCG's interview brochure says the case tests whether you can "structure a problem, prioritize and perform accurate analyses, synthesize your thoughts, and be creative" (BCG, November 2018). My scoring sheets had more rows than that, but every row fell into one of two groups: things the candidate does on a schedule I could predict before they walked in, and things the candidate does in reaction to me.
The distinction matters because of the largest meta-analysis of deliberate practice. Macnamara, Hambrick, and Oswald pooled 88 studies and found that hours of practice explained about 24 percent of the variance in performance for highly predictable activities, 12 percent for moderately predictable ones, and 4 percent for low-predictability ones. For professions as a whole the figure was under 1 percent, which is a reason to be modest about any practice plan, including this one. Reps pay best where the task is predictable, and a case interview is a mixture worth mapping.
| Scoring row | When it arrives | Predictability | Build it alone? |
|---|---|---|---|
| Opening structure | Minutes 1 to 3, every case | High | Yes; the most solo-buildable skill in the interview |
| Case math | One or two steps, usually from minute 8 | High | Yes, with an answer key |
| Chart reading | One or two exhibits | High | Yes, with an answer key |
| Brainstorming breadth | One question, mid-case | Medium | Mostly; prompts vary, the method does not |
| Synthesis and close | Last two to three minutes | High | Yes, if you record it |
| Communication | Throughout | Medium | Partly; a recording catches the fixable half |
| Reacting to redirection and pushback | Whenever the interviewer decides | Low | No; this is what the human is for |
Five of seven rows are high-predictability mechanics. That is the case for solo work. The last row is the case against pretending solo work is the whole job.
A 48-hour budget, spent two ways
Say you have six weeks at eight hours a week, so 48 hours. Two candidates spend it differently.
Candidate A follows the folklore: partner cases, as many as possible. A session runs an hour, half as candidate and half as interviewer for the other person. With scheduling friction and no-shows, 48 hours buys 25 to 40 sessions, each yielding one opening structure, one or two math steps, one close, and feedback from someone who has never sat on the scoring side of the table.
Candidate B puts 36 hours into solo reps and 12 into humans.
| Rep | Minutes each | Hours | Reps |
|---|---|---|---|
| Opening structure: read prompt, 90 seconds silent, 2 minutes spoken, 90 seconds scoring | 5 | 10 | 120 |
| Math step: spoken set-up, then check against an answer key | 4 | 6 | 90 |
| Chart read, spoken, then check | 4 | 4 | 60 |
| Sixty-second close from a written case summary | 3 | 4 | 80 |
| Full recorded case, scored the next day | 45 | 12 | 16 |
| Partner case, briefed to interrupt and redirect | 60 | 12 | 12 |
B has produced 148 opening structures to A's 25 to 40, plus 90 scored math steps and 80 scored closes that A never isolated, and B still sat 12 live cases. Redo the arithmetic with your own hours. The ratio holds at any budget, because a full case is a slow way to buy one opening.
Five reps that end in a count
Each rep below ends with something you count, not something you feel.
Opening structure
Bain's published practice cases tell candidates to take one to two minutes to think before walking the interviewer through a structure, and BCG's brochure says pausing after the question is fine. Build the rep around that. Read a prompt (McKinsey publishes practice cases with suggested answers, and any earnings headline works too), sit in silence for 90 seconds, then speak your structure for two minutes into your phone. Count three things: branches (three or four), whether each was tied to the question with a "because" or a "which would tell us," and whether you said where you would start and why.
Math
Invent a company and speak the set-up before you touch the arithmetic. Norrland Ski Rentals runs 3 shops with 120 pairs of skis each, at 60 percent utilization across a 140-day season, charging 35 euros per pair per day. Aloud: "Revenue is shops, times pairs, times utilization, times days, times price." Then: 3 times 120 is 360 pairs; 60 percent of that is 216 rented on an average day; times 140 days is 30,240 pair-days; times 35 euros is 1,058,400 euros, so about 1.06 million a season, or roughly 350,000 per shop. Score two things: did the set-up come before the numbers, and did the answer come rounded and interpreted. A wrong digit costs less on my sheet than a silent set-up followed by a correct number, because the silent version tells me nothing about how you think.
Chart
Take any exhibit from a published case. Sixty seconds: title, axes, units, then the one sentence the interviewer is waiting for: what changed, by how much, and why it matters for the question. Score: was the "so what" spoken inside the minute, and did it contain a number.
Close
Read the summary of a case you have already worked, then deliver a sixty-second close with the recorder running: recommendation first, with a number; two reasons; one risk; one next step. Score: was the recommendation the first sentence, and did it carry a number. In my experience of first rounds, roughly two candidates in eight bury the recommendation behind a recap, and the sheet punishes it every time.
The full recorded case
Once or twice a week, sit a whole case aloud, with no model answer until the recording stops. Score it the next day. Same-evening scoring is contaminated by how the case felt, and how it felt is not on the sheet.
Grading your own recording: count, do not judge
When I scored a case I was counting more than I was feeling. Countable things are hard to lie to yourself about.
| What to count | Fine | Leak |
|---|---|---|
| Seconds of silence before your structure | 60 to 120 | Under 20 (you did not think) or over 180 |
| Branches in the opening | 3 or 4, each linked to the question | 6 or more, or a list with no connecting logic |
| Math set-ups spoken before computing | Every step | Silence, then a number |
| Filler words per minute | Under 4 | 8 or more |
| Position of the recommendation in the close | First sentence | After the reasons |
| Numbers in the close | At least one | None |
Two habits make this work. Score from the recording, never from memory, because memory edits out the pauses. And score one row per listen; trying to hear everything at once is how "that felt like a seven" creeps back in.
The one thing to find a human for
The bottom row of the first table. "Suppose I told you the CFO disagrees with that." "Your second branch is not where the answer is; where would you go instead?" You cannot manufacture this alone, because an interruption you wrote yourself is not an interruption. It is also the lowest-predictability moment in the interview, which by Macnamara's numbers is where raw hours return least. That is the argument for spending human time only here, and for briefing the human.
Brief the partner before the case: interrupt my structure once with a fact that breaks one branch; challenge one number; ask "so what?" after my close. Grade only those three moments. An unbriefed partner defaults to nodding, and nodding is worth nothing.
A voice AI interviewer sits in the middle. CoachNed's live voice case interjects, redirects, and ends in a seven-score debrief, which makes it less polite than a friend but still not a human who has decided your tree is wrong and wants to watch you recover. Use it for volume and mechanics. Keep the human hours for the recovery.
Practice this today
Fifteen minutes, three reps, one recording. Pick any published case prompt. Do one opening-structure rep with the 90-second silence, one math rep with numbers you invent, and one sixty-second close. Count, do not judge.
If you want the counting done for you, CoachNed's five-minute first rep takes three typed turns of a case and scores them instantly, no account needed. The structure drill scores opening structures, the math drill covers the spoken set-up problem with an answer key, and quick math is free forever. For the full recorded case, the voice case with Ned runs about 18 minutes, and the case library has 64 original cases for when you have exhausted the firms' own. 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, or any other firm named here.
Frequently asked questions
Can you pass a case interview by practicing alone?
You can build almost every scored mechanic alone: structure, math, charts, synthesis, and most of communication. What you cannot build alone is recovery from an interviewer's redirection, so budget a handful of briefed live sessions before the real round.
How many practice cases do you need before an interview?
Fewer full cases than folklore says and far more component reps. Sixteen full recorded cases plus a hundred-odd isolated opening structures will outperform 40 unrecorded partner cases in my experience, because the sheet scores components and components are what repetition improves.
Is it better to practice case interviews alone or with a partner?
Both, at roughly three solo hours for every partner hour. Solo time builds the predictable mechanics through volume and scoring; partner time should go only to interruption, pushback, and redirection, with the partner briefed to supply them.
How do you practice case interview math by yourself?
Speak the set-up before computing, then check against an answer key. The set-up is what the interviewer scores; the arithmetic is only evidence that the set-up was right.
Sources
- The role of deliberate practice in the acquisition of expert performance (Ericsson, Krampe, and Tesch-Römer, Psychological Review, 1993) — the feedback and repetition conditions, and the violinists' rating of practice alone. Checked 2026-09-24.
- Deliberate practice and performance in music, games, sports, education, and professions: a meta-analysis (Macnamara, Hambrick, and Oswald, Psychological Science, 2014) — variance explained by practice, by domain and by task predictability. Checked 2026-09-24.
- Test-enhanced learning (Roediger and Karpicke, Psychological Science, 2006) — rereading versus self-testing, 40 percent versus 61 percent recall after one week, and the confidence gap. Checked 2026-09-24.
- BCG interview preparation brochure (November 2018) — what the case tests, and that pausing to structure is fine. Checked 2026-09-24.
- BCG case interview preparation — BCG's current description of the case interview. Checked 2026-09-24.
- Bain practice case: Coffee Shop Co. — the one-to-two-minute thinking pause before structuring. Checked 2026-09-24.
- Interviewing at McKinsey — McKinsey's published practice cases with suggested answers. Checked 2026-09-24.