Framework Mistakes 2026: The Six Named Failure Modes

The six case interview criteria a firm actually publishes

Oliver Wyman publishes six evaluation questions for its case interviews, each with the reason it asks. They cover how a candidate approaches an unstructured problem, how they evaluate and use data, whether they can think beyond the obvious, whether they apply common sense, whether they ask the right questions, and how they communicate their ideas, checked 30 September 2026 at oliverwyman.com.

None of the six mentions a framework. That absence is informative, because this is the closest thing to a published marking scheme that any firm offers, and the word does not appear in it.

The same page describes the relationship it expects: "Think of your case interviewer as your client. They want you to solve the problem, and can help you to do so. Work together." Bain puts it as a "structured conversation" and says it is "not looking for a single right answer", checked 29 September 2026 at bain.com.

Read those criteria as a list of things a bad structure prevents. A grid recalled in advance cannot demonstrate how you approach an unstructured problem, because you did not approach it. It cannot demonstrate that you asked the right questions, because the questions come from the grid rather than from the problem.

Naming the six framework mistakes

Each of these is a distinct mechanism with a distinct fix, which is why they are worth separating rather than lumping together as a lack of adaptation.

1. Choosing before reading. The tool is selected from the prompt's industry rather than from its question, so the structure answers a different question correctly.
2. Overlap. Two branches contain the same item, so no total closes and every saving is overstated.
3. The wrong base. A percentage is applied to revenue, cost or profit interchangeably, and the three are different numbers.
4. Category depth. The tree reaches four levels of nouns and never reaches a measurable quantity.
5. Budget burn. So much time goes into producing the structure that nothing is left for the analysis it was supposed to organize.
6. Holding past the ceiling. A branch is continued after the arithmetic shows it cannot contain the answer. When to use a framework is about this one specifically.

Three of the six are arithmetic and can be checked on the page in front of you. The sections below price them.

The overlap that breaks a total

A distributor has 88 million dollars of annual cost and wants 10 percent out of it. The candidate builds five cost buckets from a supply chain grid.

The buckets sum to 93 million against an 88 million dollar base, which is 5.7 percent too much. The cause is 5 million dollars of inbound freight that the client books inside landed product cost and the grid also books under logistics. It is in procurement and in logistics at once.

Now watch what the overlap does to a recommendation. A 12 percent saving on logistics of 14 million dollars is 1.68 million. The same 12 percent on the true logistics figure of 9 million is 1.08 million. The promise is overstated by 600,000 dollars, which is 1.68 divided by 1.08, a factor of 1.56.

The fix is the sum check, and it takes ten seconds. Add the buckets and compare them with the parent. A cut that overshoots has an overlap and a cut that undershoots has a gap, and either one is worth finding before the recommendation is built on it. The MECE principle covers the recut.

The wrong base, a framework mistake priced three ways

Same distributor: 104 million dollars of revenue, 88 million of cost, so 16 million of operating profit, which is a 15.4 percent margin. A candidate says the initiative delivers a 2 percent improvement. Three bases, three answers.

The spread between the largest and the smallest is 2.08 divided by 0.32, which is 6.5 times. The gap between them is 1.76 million dollars, which is 5.5 times the smallest of the three figures. A finance director hearing 2 percent will assume one of these and the candidate will mean another.

The habit that prevents it is one word long. Never say a percentage without the noun it is a percentage of: 2 percent of cost, two points of margin, 2 percent of profit. Charles River Associates asks candidates to "show your math", checked 30 September 2026 at crai.com, and the base is the part of the math that most often goes unwritten.

The same error has a second life at the end of a case, in the summary. A number computed on one base and then compared with a number computed on another produces a sentence that is arithmetically sound and factually wrong, which is the hardest kind of error to catch.

Budget burn, counted in minutes

A structure that consumes the case is a failure even when it is a good structure. Count the minutes on a thirty-minute case.

Recalling a named grid, writing out its four or five boxes legibly, and reading it back aloud takes about four and a half minutes at interview pace. The grid then needs filling, and because it was not derived from the prompt it generates clarifying questions: three of them at roughly forty seconds each is two minutes. That is six and a half minutes, which is 21.7 percent of the case, spent before a single quantity has been estimated.

Deloitte names the symptom directly: "Ask questions, but don't use them as a crutch. While it's important to ask clarifying questions, avoid relying on them too heavily. We're interested in how you connect the dots when solving the problem", checked 30 September 2026 at deloitte.com. An unfilled grid is a question generator, and the questions it generates are about the grid.

A derived structure costs less because it asks for fewer things. Writing the identity for the quantity in the prompt takes ninety seconds and tells you exactly which three numbers to request. Stated in minutes the comparison is six and a half against two, which is the part that decides how much case is left.

The case interview mistake that survives to the recommendation

Five of the six failure modes are visible while they happen. The sixth is not, and it is the one that costs offers: a correct calculation described by a wrong sentence.

The overlap example above produces one. A candidate computes 1.68 million dollars correctly from the stated 14 million and then says the logistics branch delivers most of the ten percent target. Ten percent of 88 million is 8.8 million, so 1.68 million is 19 percent of the target, and the true figure of 1.08 million is 12 percent. The arithmetic was right and the word most was describing 12 percent of the target.

Guard against it with a two-part check at the end of every calculation. First, does the number land where I expect. Second, does the sentence I am about to say describe what the number shows. A share of a target, a share of profit and a share of a gap are three different claims, and only one of them is usually true.

Oliver Wyman's published method ends with making a recommendation, and its earlier step asks candidates to use numbers. Between those two steps sits the translation, which is where this failure lives.

Where each mistake lands on the rubric

The weights below are ours, read from the rubric code our own grader runs. No consulting firm publishes a marking scheme, so none of this is Oliver Wyman's reading of its own six criteria.