The number you land on is rarely the first thing on the scoring sheet. In a live case, I am grading whether I can audit your math while you do it: equation stated before you multiply, units tagged, rounding error named, and a recovery that does not burn the calendar when you slip.
That is the thesis most candidates miss. They practice speed drills as if the room were a timed quiz. Firms design the live case as a conversation about how you think with numbers, which is why Bain tells candidates to "think aloud" and share calculations even when unsure, and why BCG frames the case as structuring, questioning, analyzing data, and "perform[ing] quick calculations" while you communicate. After this piece you should be able to run a scored math minute the way an interviewer listens: setup first, then arithmetic, then meaning.
This guide is about behavior in the room. Mental shortcuts live elsewhere; formulas live elsewhere; the practice program lives elsewhere. Here we stay on what the sheet rewards when the clock is running and the calculator is gone.
What the sheet actually rewards
In a typical first-round slate of eight candidates, I expect two or three who can multiply under pressure and still lose the math marks because the interviewer never saw the model. Rough estimate from years of scoring, not a published firm statistic.
| Scoring signal | What I hear | What I mark |
|---|---|---|
| Setup | "Incremental revenue equals price times volume delta" | Structure and logic |
| Units | "Annual dollars, not monthly" | Care / error control |
| Rounding | "I'll use $48M, call it roughly 4% light" | Judgment |
| Sanity check | "Too big vs. total revenue of $120M" | Business sense |
| Recovery | "I mis-placed a zero; correcting to…" | Coachability |
McKinsey describes the problem-solving interview as a discussion of a typical client scenario to evaluate analytical thinking and approach, not a race to a hidden answer (Interviewing at McKinsey). Bain's sample FashionCo case walks candidates through revenue options with labeled drivers and a clear comparison, which is the same pattern: the math exists to decide between choices, not to decorate the whiteboard (FashionCo.).
Ned's rule. If I cannot reconstruct your equation from what you said in the first fifteen seconds, the rest of the arithmetic is noise on my sheet.
Calculators: what the firms actually say
Treat live case math as pen-and-paper with a spoken audit trail. Bain states this plainly on its digital-assessment FAQ: for Sova and TestGorilla, "a pen, paper and calculator are allowed," and then the critical clause — calculators are allowed for the digital assessment only, "not during any subsequent interviews" (Digital assessment). That is firm language, not candidate folklore.
Online screens are a different product. McKinsey's Solve assessment prohibits applications, websites, AI tools, and pre-prepared notes unless you have an approved accommodation (Solve). BCG's public case-prep page emphasizes quick calculations inside a spoken problem-solving conversation (Case Interview Preparation). Prep sites invent calculator myths; when firms speak, they either ban the tool for the live interview (Bain) or describe skills, not gadgets (BCG, McKinsey interviewing pages).
Practical implication: practice the room standard — no calculator — even if an online screen later lets you use one. The scoring behavior below is what transfers.
Set up the equation aloud before you compute
Say the model before any digit leaves your mouth.
Bad: "So uh 47 times 1.2 is… wait… 56.4… no…"
Good: "I need year-one contribution from the loyalty program: participating customers times spend times (1 minus discount), minus the activation fee per participant. Inputs: 10 million customers, 25% join, $100 spend, 20% off, $50 fee."
That second version is already half-scored even if the multiply later wobbles. Bain's tip list puts structure and thinking aloud ahead of polish (Case interview preparation). If the setup is wrong, a good interviewer redirects you before you waste three minutes; if you never state it, they cannot help.
Script I want to hear every time:
- Unknown: what decision number are we producing?
- Equation in words: drivers and operators.
- Inputs with units: dollars per year, percent, headcount.
- Then: "I'll compute."
Silence during arithmetic is fine. Silence during setup is not.
Round with a stated error band
Rounding without narration looks like sloppiness. Rounding with an error band looks like judgment.
Say: "I'll take 47% as 50%, so I'm high by about three percentage points; if the decision is close I'll tighten." Or: "$312 million as $300 million — roughly 4% low."
| Original | Rounded | Stated band | When to tighten |
|---|---|---|---|
| 47% | 50% | ~3 pp high | Margin vs. target under 5 pp |
| $312M | $300M | ~4% low | Payback near the hurdle |
| 4.7 years | 5 years | ~6% long | NPV sign flip risk |
I stop listening when candidates apologize for rounding instead of owning the band. Firms want directionally correct decisions under time; Bain's coffee-shop and FashionCo materials reward clear comparisons, not six-decimal theater. Round for direction, name the error, tighten only if the call flips.
Units and time frames are where most silent fails live
The arithmetic was right; the unit was wrong. That still fails.
Tag every input as you write it: $ / customer / year, customers, % of customers, $ one-time. Say the time frame out loud when you mix annual revenue with a monthly cost: "I'll annualize the $2M monthly opex to $24M."
Common traps I see in the room:
- Treating a one-time fee as annual (or the reverse).
- Mixing percent-of-revenue with percent-of-customers without converting.
- Using U.S. population for a city market, or vice versa, without saying so.
- Quoting "million" when the pad says "billion."
If you catch a unit mismatch mid-calc, stop and restate. That pause scores better than a confident wrong answer.
Sanity checks before you announce the number
Before you speak the result, run one order-of-magnitude check against a known anchor from the case.
Example: client revenue is $120M. You just computed a $90M incremental from a 20% discount on half the base. That is almost another full business. Either the setup is wrong or the insight is "this lever is enormous — confirm volume assumptions." Both are useful; a naked "$90M" with no check is not.
Three checks that take under ten seconds:
- Compare to total revenue, profit, or market size already on the page.
- Flip the percent: "is 3% of $400M about $12M?"
- Ask whether the sign matches the story (cost cut should raise margin, not lower it).
Ask for a number versus assume
Candidates either freeze waiting for permission or invent three decimals of fiction. Use a decision rule.
| Situation | Ask | Assume |
|---|---|---|
| Driver will flip the recommendation | Yes — one precise ask | — |
| Missing a second-order input | — | State a round base case and move |
| Interviewer already gave the figure | Confirm once, then use it | — |
| Market-sizing with no data pack | — | Segment, state assumptions, invite correction |
Phrase assumptions as constraints, not apologies: "I'll assume 20% of households buy at $40; if that's high, the market shrinks linearly." Bain explicitly encourages clarifying questions when needed and adapting to cues (Case interview preparation). One clean ask beats a fishing expedition.
Recovering from a slip
Everyone slips. The score separates candidates who freeze from candidates who correct.
Protocol:
- Stop the wrong branch out loud: "I see an error — I used monthly cost as annual."
- State the fix in one line.
- Recompute only the broken step, not the entire tree unless you must.
- Re-interpret: what changes for the recommendation?
Do not apologize three times. Do not erase the pad into illegibility. Interviewers mark coachability here; Bain's "listen and adapt" tip is the same idea in softer language.
Before / after: one transcript, one decision
Client: regional cafeteria supplier, $80M revenue, considering a $4M annual contract with a university. Variable cost is 60% of revenue. Question: does the contract clear a $1.2M contribution target?
Before (loses math marks):
"Okay so 4 million times… 40 percent is… 1.6? Wait, 60 percent cost so margin is 40… 4 times 0.4 is 1.6 million, yeah that's above 1.2 so yes."
What I heard: no equation named, units implied, no check against revenue, no stated rounding, recommendation without saying contribution versus revenue. If the 60% was cost of goods only and opex still hits the contract, the candidate never opened that door.
After (scores):
"I'm solving for annual contribution: contract revenue minus variable cost. Equation: $4M times (1 − 0.60) = $4M × 0.40. That's $1.6M contribution. Sanity check: $1.6M is 2% of the $80M base business, and it's $0.4M above the $1.2M target — so yes on contribution alone. Caveat: I'm treating the 60% as fully variable on this contract and ignoring incremental fixed cost; if the kitchen needs a new line, I'd subtract that next."
Same arithmetic. Different score. The second version is auditable, tagged, checked, and decision-ready.
Reproduce it: $4,000,000 × 0.40 = $1,600,000; gap to target = $400,000; share of firm revenue ≈ 1.6 / 80 = 2%.
Practice this today
Run one math minute out loud with a timer set to 90 seconds. Use only pen and paper. Force the sequence: unknown → equation → units → round with band → compute → sanity check → interpret. If you slip, use the recovery protocol instead of restarting.
Then do three graded reps on CoachNed's free quick-math drill, or open a full voice case on /interview and treat every quantitative turn the same way. Everything is open for seven days, no card; then $120 for a recruiting season or $49 a month.
Frequently asked questions
Are calculators allowed in McKinsey, BCG, or Bain case interviews?
For Bain, the firm states calculators are allowed on the digital assessment only, not in subsequent interviews. McKinsey and BCG describe live cases as problem-solving conversations with quick calculations; treat them as no-calculator unless your recruiter writes otherwise. Online screens can differ — follow the invitation email.
Should I ask permission to round in a case interview?
You can, once, early: "I'll round as I go and call out the error band." Most interviewers nod. The scoring win is naming the band, not the permission ritual.
What do I say if I realize my math is wrong mid-answer?
Stop, name the bug, correct the broken step, restate the implication. Do not restart from zero unless the whole tree depended on the error.
How precise does case interview math need to be?
Precise enough that the recommendation does not flip inside your stated error band. Directional rounding is fine when labeled; fake precision without a check is not.
Is case math mostly about speed?
Speed matters only after the setup is clear. I would rather hear a clean 70-second audit than a 25-second black-box answer I cannot trust with a client.
Sources
- Interviewing at McKinsey — problem-solving interview as analytical approach on a client scenario. Checked 2026-09-24.
- Solve, McKinsey's assessment game — integrity rules for the online assessment (no apps, websites, AI tools, or pre-prepared notes). Checked 2026-09-24.
- BCG Case Interview Preparation — case includes data analysis and quick calculations with clear communication. Checked 2026-09-24.
- Bain digital assessment — calculator allowed on Sova/TestGorilla only; not in subsequent interviews. Checked 2026-09-24.
- Bain case interview preparation — think aloud; share reasoning and calculations; clarify and adapt. Checked 2026-09-24.
- Bain FashionCo. case study — quantitative comparison of revenue options with labeled drivers. Checked 2026-09-24.
CoachNed is not affiliated with McKinsey, BCG, Bain, Deloitte, or any firm named above.