How Case Interviews Are Scored: The Published Evidence

A case interview is scored question by question, and a firm shows it

Most advice treats the case interview as one graded event with one standard. The only firm that publishes criteria at the question level contradicts that.

The Dairy Farm case sits at oliverwyman.com, labeled a practice example at medium difficulty. A private equity client has to decide whether to invest in one of two dairy farms in an Illinois county. Under the opening question, the firm lists what it wants: the key question identified, the relevant information picked out, a clear structure, and that structure communicated succinctly. Under the next question the list changes. Now it wants several factors produced, adaptation as the data arrives, and the data laid out so it can be computed quickly.

Read those two lists in sequence and the mechanism becomes visible. In the opening, brevity is credited. Two minutes later, breadth is credited. A candidate who keeps optimizing for succinctness through the middle of the case is answering the first standard while the second one is being applied.

The published model answer has three branches: farm profitability, local dairy market dynamics, and fit with the client's portfolio. That third branch is worth noting on its own, because it is the branch that most candidates skip. The client is a fund. A fund does not buy a farm in isolation, it buys an asset into a portfolio, and the published answer treats that as a first-level branch rather than a footnote.

Every stage eliminates, which changes what a score is for

A score only matters if it decides something. One regional page says exactly what it decides.

Oliver Wyman's page for Brazil and Mexico, at oliverwyman.com, publishes four stages: application, online tests, a fit interview, then case interviews. It states that "all stages are evaluative and eliminatory". Then it adds the sentence that makes the stakes concrete: "If your application is unsuccessful at the fit interview stage, you'll need to wait one year before reapplying."

The same page publishes the case volume. Two rounds, two interviews per round, four cases in total, each with its own case. The fit interview runs about fifteen minutes, in English, with no case to solve.

These numbers are regional. They cover Brazil and Mexico, and the firm's global preparation page publishes no round count at all. Presenting them as the firm's worldwide process would repeat an error this site has already made and corrected.

What travels beyond the region is the structure of the claim. Each stage carries its own decision, so a strong case in round two cannot repair a weak fit interview in stage three, because stage three already closed. That is why the fifteen-minute conversation with no case in it is the most expensive fifteen minutes in the published process, and why its content is the biggest hole in the public record.

The two tiers inside a single published case interview answer

Two firms publish their own cases with two graded answers per question, and the gap between the tiers is the same in both.

Deloitte's LeadAuto case, extracted from the page data at usrecruiting.deloitte.com, separates "A good answer may include" from "In addition to the above, a great answer may also include". For its first question, the upper tier adds a MECE internal and external split, a multi-year contract commitment, a joint venture, expansion beyond Asia, and higher-margin services.

Bain runs the same device across both of its home-grown cases, labeling the tiers "A good answer" and "A better answer".

Line the two up and the upper tier holds the same four kinds of content in both firms: an alternative the client could take instead, a risk attached to the recommendation, a time horizon, and a next step. Neither firm says that is the rule. Two instances are two instances, and we report them as that rather than as a market pattern.

The operational value is a test you can run on yourself in ten seconds. Write your recommendation. Then check it for an alternative, a risk, a horizon and a next step. A recommendation missing all four is the lower tier of two firms' own published ladders, which is a more useful diagnosis than a feeling that the close was weak.

Our weights, and the arithmetic they protect

The weights below are ours. They come from the grading code in src/lib/grading, and no consulting firm publishes a scheme of its own, in this corpus or anywhere we have looked.

What case interview feedback can and cannot tell you

One firm promises to tell you the result immediately, and the same firm declines to describe how it was produced.

Bain writes at bain.com that "any feedback you receive will be based on an objective scoring system" and that "After each interview, you'll learn exactly how it went." That is the only reference to a scoring system in our entire official corpus, across forty-four firm files. Nothing else is published: no dimensions, no scale, no aggregation, no threshold.

So a feedback conversation gives you an outcome and a narrative. It does not give you a breakdown you can practice against, because the breakdown stays inside the firm.

Three things follow, and they are practical.

1. Write down the feedback in the words you heard it. Paraphrasing it into rubric language invents precision that was never there.
2. Separate the diagnosis from the example. If an interviewer says the structure was thin and cites the cost side, the diagnosis is coverage and the example is cost. Practicing cost trees fixes one case. Practicing coverage fixes the category.
3. Do not infer a threshold. Nobody publishes one, and a candidate who decides they were two points short is describing a number that does not exist.

Building your own case interview scorecard from what is published

You can assemble a workable self-assessment entirely from firm pages, at no cost, in one sitting.

1. Take the two Oliver Wyman question-level lists as a timeline. Grade your opening on key question, relevant information, clear structure and succinct delivery. Grade your middle on factors produced, adaptation to new data, and data laid out for fast computation.
2. Take Accenture's published sentence as your four analytical lines. Problem definition, logic of structure, quantitative skill, communication. The workbook states that success "will not depend on identifying the "correct" answer", and it also prints worked answers whose numbers are right, so treat a correct number as a floor and the reasoning as the score.
3. Take the Bain and Deloitte upper tier as your close. Alternative, risk, horizon, next step.
4. Add Poise from Accenture's soft evaluative list and leave it ungraded. No text tool scores composure, and pretending otherwise would be worse than admitting the gap.

That is fourteen items across four checkpoints, all attributable, none invented.

Where our grader stops

Our math drill returns Answer, Working, Units and Interpretation as four separate lines, and our chart drill returns So what, Observation, Implication and Next test. Both are free, and neither asks for an account. The reasoning behind every weight sits on how an answer is assessed.

Both tools stop in the same place. They read what you typed. They cannot hear a pause that lasted too long, they cannot tell whether a partner would have trusted you in front of a client, and they will not choose a recommendation for you. The firms' own published criteria include items of exactly that kind, which is the honest reason to run drills for the mechanics and a person for the rest.

There is a second limit worth naming. A drill grades one artifact at a time, and a case interview is graded as a sequence. The Dairy Farm criteria show why that matters: the standard applied to your opening is gone by the time you are laying out data for a calculation, and no single-artifact score captures the handoff between the two. Running four drills back to back gets closer to that, and it still is not the same thing as one continuous conversation with a person who remembers what you said four minutes ago.

For the criteria lists themselves, set out firm by firm, see what interviewers score. For the behavior the same pages ask for during the conversation, see communicating in a case. For the two firms' published cases with their full arithmetic, see case interview examples.