CoachNed
Two empty chairs across a long interview table

Case interview prep

Case Interviews as a Non-Native English Speaker: What Costs Points

Interviewers do not score accent. They score numbers they can hear and structure they can follow. A working-memory plan for casing in your second language.

UpdatedReviewed by Ned

The scoring sheet I filled in after several hundred case interviews had no row for English. It had rows for structure, numbers, judgment, and communication, and a non-native speaker can leak points in all four without making one grammar mistake. The leak is not accent. It is working memory: a case asks you to hold a tree, a running number, and the interviewer's last sentence in your head at once, and a second language taxes exactly that capacity.

So the fixes most candidates buy, a bigger vocabulary or a lighter accent, aim at the wrong target. The fix that works is to move load off your tongue and onto the page, then spend the capacity you free on the two things the sheet rewards most: numbers the interviewer can repeat back, and structure the interviewer can follow.

By the end you will have a number protocol borrowed from aviation, an eight-phrase card for the connective language of a case, and a worked example you can score yourself.

What the sheet has no row for

Most firms score a case on four to six dimensions: structure, numbers, business judgment, and whether a client could have followed you. Nobody scores pronunciation. On my own sheets I have written "lost the number" and "structure disappeared inside the sentence" many times. Never "accent."

Language still costs you, and the mechanism tells you where to spend effort. In 2010 Shiri Lev-Ari and Boaz Keysar had native English listeners rate the truth of trivia statements read aloud by native and non-native speakers; the listeners knew the speakers were only reciting identical statements. On a 14-centimeter truth scale, native speakers scored 7.59; the same statements scored 6.95 with a mild accent and 6.84 with a heavy one. The authors' explanation is processing fluency, not prejudice: speech that is harder to process feels less true, and listeners blame the content.

The second experiment is the useful one. When listeners were warned that difficulty might bias them, the mild-accent penalty vanished (7.52 against 7.47) while the heavy-accent penalty stayed (6.9). Your interviewer is a forewarned listener trying to be fair; the correction works when your speech is easy to process and fails when it is not. So the game is making the parts that carry points, the numbers and the structure, easy to process. A small study, under thirty listeners per experiment; the sizes are directional, the direction is not.

The three leaks, and where each one lands

LeakWhat I experience from the chairWhere it landsThe fix
Numbers I cannot repeat back: fifteen or fifty, million or billion, a swallowed decimalI stop following your logic and start reconstructing the figureNumbers, then credibility on everything afterThe readback protocol below
Sentences that outlive their structure: verb at the end, bucket label buried in the middleI lose the tree while waiting for the verbStructure and communicationHeadline first, one clause per sentence
The repair loop: restarts, apologies for your English, narrating your difficultyI start scoring your confidence instead of your answerCommunication, plus the minute it costA scripted recovery line, never an apology

Notice what is missing. Grammar. A wrong article, a missing third-person "s", a preposition imported from your first language: I have never seen one of these move a score when the sentence was short and the number was right. Candidates spend the week before a first round on the thing that costs nothing.

Move the load onto the page

Working memory is the scarce resource in a case for everyone, and for you part of it is busy assembling English. The answer is not to try harder. It is to store less in your head.

Write more than a native speaker would. Full bucket labels, not abbreviations you will have to expand out loud. The formula before you say a word of it. The three lines of your synthesis, then read them. Reading from your page is not a weakness; it is the format.

Saying less is free. Slow from a nervous 170 words a minute to a deliberate 120 across twelve minutes of talking, my estimate of your share of a 25-minute case, and you drop about 600 words. They are the ones carrying the restarts and the apologies. No candidate has ever failed a case for saying 600 fewer words. Several have failed for saying them.

The connective tissue of a case, the openings, transitions, and closes, is where second-language speakers improvise badly and burn capacity. Script it. Each phrase below is under ten words, puts the verb early, and has no subordinate clause.

MomentSay thisWhy it works
Presenting structure"Two branches: demand, then cost. Demand first."Labels before detail
Moving on"That closes demand. Now cost."One clause; the interviewer can steer
Before math"I will write it, then say it."A pause without an apology
Saying a result"Forty-three million. Four-three."Digits kill the teen/ty ambiguity
Missing a figure"Was that fifteen or fifty?"A precise request reads as rigor
A word fails"The fee the platform keeps. I will call it take rate."Define, name, move
Taking pushback"The number holds. The assumption may not."Separates math from judgment
Synthesis"Recommendation first, then two reasons, then one risk."Announces its own structure

Memorize them in your own accent; when a phrase is automatic, the working memory that would have built it goes to the case.

Say numbers the way pilots do

Aviation solved the teen/ty problem decades ago, because fifteen and fifty sound alike on a radio and the stakes are altitudes. The FAA's Aeronautical Information Manual requires most numbers to be spoken digit by digit, so 13,500 feet is "one three thousand five hundred", and clearances are read back so both sides heard the same figure. That is the standard a case interviewer needs, and almost nobody applies it.

  1. Round before you speak. 43.2 becomes "about 43." The precision was never going to be scored; the audibility is.
  2. Number, unit, digits. "Forty-three million. Four-three." For any figure ending in -teen or -ty, add the digits. Say "percent" in full every time.
  3. Stop after the unit. Let the interviewer write. The silence you want to fill with "so, this means that, basically" is the moment they catch up with you.

Ned's rule. I never mark down an accent. I mark down every number I have to ask for twice, and every tree I lost inside a sentence. Say the number, say the unit, stop talking.

A worked example you can score

Marram Ferries, an invented Baltic operator, has revenue of 240 million euros and operating profit of 31 million. Fuel is 18 percent of revenue. What happens to profit if fuel prices rise 25 percent? Fuel is 43.2 million; a quarter more is 10.8 million; profit falls from 31 to 20.2 million, roughly a third.

The delivery I hear most often under pressure, 73 words:

"So if we take the revenue, which is two hundred forty, and the fuel is eighteen percent of this, so it is, sorry, forty-three point two, and then if it increases with twenty-five percent this would be, so this is ten point eight more, so the profit which was thirty-one it becomes now, sorry, twenty point two, so it is decreasing with I think thirty-five percent approximately, if my calculation is correct, sorry."

The same arithmetic at readback pace, 45 words:

"Fuel today: 18 percent of 240 million. That is 43 million. Four-three. (pause) A 25 percent rise adds a quarter of 43, about 11 million. (pause) Profit falls from 31 million to about 20 million. Two-zero. That is a third of profit gone. Fuel is the case."

What I am scoringFirst deliverySecond delivery
Numbers I could write down without asking2 of 5; the decimals blurred past5 of 5
Structure signaled before contentNone; one continuous clauseThree steps, each announced
Confidence signalsThree apologies, "I think," "if my calculation is correct"None

No word in the second delivery is beyond an intermediate learner; the difference is that the second candidate wrote three lines and read them.

Which language the interview will be in

Requirements are office-specific and they move; treat everything below as "as of September 2026" and confirm with your recruiter in writing.

BCG is the most explicit. Its Switzerland FAQ states that interviews are generally conducted in German or French, with at least one interview in English, and that candidates not fluent in German or French are interviewed entirely in English. Its campus pages carry a worldwide list of office language requirements in three tiers: English only (London, the US offices, the Nordics, Singapore, among others), local language recommended (Amsterdam, Geneva, Zurich, Montreal), and local language required (Germany, France, Italy, Spain, Japan, Korea, Greater China, Brazil, Mexico, Poland, and more). Two of BCG's own pages disagree about Athens and Tokyo; weigh any list accordingly.

McKinsey states requirements office by office. The Finland page says Finnish is not required but an excellent command of English is; the German-language recruiting site says plainly that the one-to-one interviews take place in German. Bain's 2026 Bainworks page lists C1 in both French and English for France, Dutch for the Netherlands, and Spanish for Spain.

What is folklore: "the case is always in English and the fit interview is always in the local language." The pattern exists, but it varies by office, role, and year. Ask the recruiter in writing which interviews will be in which language, and whether that covers both the case and the personal-experience part. Then prepare in exactly those languages, including the math.

Fit stories and pushback in a second language

Fit stories feel as if they need idiom. They do not. The interviewer is listening for a decision, a number, and what you did next. Translate those three things and let the color go. A story that only works with an untranslatable joke is the wrong story.

Pushback is not a comment on your English. When an interviewer says "I am not sure I agree," second-language candidates often hear "I did not understand you" and restart the whole answer, slower and longer. Do the opposite: split the challenge into the number and the assumption, defend whichever holds, concede whichever does not, in two sentences.

One finding runs the other way. In 2012 Keysar, Hayakawa, and An found that people choosing in a foreign language were not swayed by framing and showed less loss aversion than people choosing in their native tongue; the published paper attributes this to the cognitive and emotional distance a foreign language creates. One lab finding is not a law of interviewing. But the judgment row rewards exactly that distance, and I have watched more than one second-language candidate give the calmest recommendation in the round.

Practice this today

One assignment, forty minutes. Take the math block from any case you have done and record it twice: once at your normal pace, once at readback pace with the number-unit-digits protocol and a written line for each step. Play both back and count numbers a stranger could write down, restarts, and apologies. If the second recording does not win on all three, you are still computing and speaking in the same breath.

Then do one typed rep at /start: three typed turns of a case with instant scores, no account needed. Typing removes pronunciation, so a low structure score there means the leak is structure, not speech. After that, run the live voice case with Ned at /interview; the seven-score debrief separates communication from numbers, which is this article's diagnosis. The math drill gives you fresh figures to read back out loud, and the quick-math drill is free forever.

Practice

Math drill

Answer a real case-math prompt and get AI-scored feedback on setup, units, and the business meaning.

Start a drill

Everything on CoachNed is open for seven days, no card; then $120 for a recruiting season or $49 a month. CoachNed is independent and is not affiliated with McKinsey, BCG, Bain, or any other firm named here.

Frequently asked questions

Will a strong accent fail me in a case interview?

Not on its own. No major firm publishes a pronunciation standard; interviewers score structure, numbers, judgment, and communication. An accent costs you when it makes a number or a bucket label hard to catch, so the fix is slower numbers with digits, not accent reduction.

Should I ask the interviewer to repeat a number?

Aviation solved the teen/ty problem decades ago, because fifteen and fifty sound alike on a radio and the stakes are altitudes. The FAA's Aeronautical Information Manual requires most numbers to be spoken digit by digit, so 13,500 feet is "one three thousand five hundred", and clearances are read back so both sides heard the same figure. That is the standard a case interviewer needs, and almost nobody applies it.

Yes, once and precisely: "Was that fifteen or fifty?" A precise request reads as rigor. A vague "sorry?" three times, or ten minutes of math on a guessed figure, does not.?

Notice what is missing. Grammar. A wrong article, a missing third-person "s", a preposition imported from your first language: I have never seen one of these move a score when the sentence was short and the number was right.

Is it better to interview in my native language if the office offers both?

Interview in the language your casing vocabulary lives in. If you learned to case in English, an English case will be stronger even in a local-language office. If you are equally strong in both, the local language answers a question the interviewer has anyway.

Sources