What to take away
- A structure is a plan for finding the answer, not a checklist you display to prove you prepared
- Framing is the sub-skill that marks your structure, and it asks one question: does this fit this problem
- A fitted three-branch tree outscores a canonical five-box framework almost every time
- The tell of a memorised framework is that it would fit any case in the book, which means it fits none of them well
- Framework patterns are useful as equations and decision criteria, so borrow the logic and drop the boxes
- Interviewers mark the structure you commit to, so redraw a bad cut out loud rather than defending it
What is a case interview framework actually for?
A framework is a plan for finding the answer. That is the whole job. Before you touch a number or open an exhibit, you say how you intend to break the problem into parts, which part you think holds the answer, and where you will look first. The interviewer writes down whether that plan was any good.
Most preparation quietly replaces that job with a different one. Frameworks get learned as objects to be shown: four boxes on the page, delivered in order, proof that the reading was done. The structure becomes a performance rather than a tool, and the difference is audible within ten seconds.
What the interviewer is watching in that minute
The structuring moment is the most diagnostic part of a case, because it is the only stretch where nobody is leading you. Later there are exhibits to react to and questions to answer. Here there is a vague commercial problem and a blank page, and your job is to carve it. So the interviewer is not checking your structure against a model answer. They are asking whether the cut was intelligent, whether the pieces stay separate, and whether you had a view about which piece mattered.
On our rubric this sits inside analytical thinking, which carries 50% of the score, and the sub-skill is framing. It is marked explicitly when you present the structure, then implicitly for the rest of the case, because every later move either uses the structure or quietly abandons it. Abandoning it costs the mark twice: once on framing, and again on a closing synthesis that connects to nothing. Eight minutes in, when a new number arrives, you should still be able to say which branch it lands in.
Why do memorised frameworks lose points?
Because of what framing marks. The anchor for a strong score is a structure that fits this problem, broken into parts that do not overlap, with a stated hypothesis or a named place to start. Notice what is missing: nowhere does it credit completeness, vocabulary, or the number of boxes. A memorised framework is optimised for the things nobody is marking.
So a three-branch tree built in the room beats a five-box framework carried in from outside, and it is not close. The fitted tree earns the fit, and usually the hypothesis too, because you chose the branches and therefore have opinions about them. The canonical framework earns coverage, which nobody is counting, then costs you when you are asked why the second box matters here.
Recitation is audible
Interviewers hear the same dozen structures every recruiting season, so recognition takes no effort. The cadence smooths out and speeds up, because the words are being retrieved rather than composed. The nouns go generic: market, competition, capabilities, and nothing from the prompt. And one branch usually has no business being there, kept because the framework includes it, and it is the one the candidate cannot defend.
None of that is punished for being unoriginal. It is punished as evidence: somebody who applies a template to an unfamiliar problem in an interview will do it on a client engagement too.
Structure B is not cleverer than Structure A. It is the profit tree, cut where this problem happens to be interesting. That is what the patterns are for.
Which framework patterns are worth knowing?
Six, rather than the forty in the average framework list. Each is really an equation or a decision criterion with a habit attached. Know them the way you know that revenue is price times volume: as something you reach for without noticing.
| Pattern | The question it fits | The trap |
|---|---|---|
| Profit tree | Profit has fallen, or has not risen when it should have | Drawing both sides at equal depth when the prompt has already told you which side moved. If volume rose while profit stayed flat, price and cost to serve deserve the detail |
| Market entry | Should the client enter this market, and how | Four boxes of market, competition, capabilities and entry mode with no decision criterion. Entry is a decision, so it needs a number the client would accept: profit by year three, or the share needed to break even |
| Deal or M&A screen | Is this target worth buying at this price, by this buyer | Writing a market-attractiveness essay and forgetting the buyer. A screen needs the target's standalone economics, what this owner adds that another would not, and the exit |
| Pricing | What should this cost, or why is our price wrong | Naming cost-plus, competitor-based and value-based as though listing the three methods were an answer. Pick one as primary and say why: cost is the floor, willingness to pay the ceiling |
| Growth | Revenue has to grow by a stated amount, from somewhere | A grid of organic against inorganic with no view of the size of the gap. Size the gap first in the client's own units, then cut by the sources that could close it in the time available |
| Capacity and operations | Throughput is too low, or cost per unit too high, in a process you can walk | Reaching for efficiency ideas before naming the constraint. A process has one binding bottleneck at a time, and a structure that does not hunt for it finds savings that raise output by nothing |
What these share matters more than the individual boxes. Each starts from something arithmetic or something decidable: the profit tree from an identity, pricing from a floor and a ceiling, entry and deal screens from a criterion the client would sign off on, growth from a gap you can size. That is the transferable part.
How do you build a fitted structure in ninety seconds?
Ask for the time first. Sixty to ninety seconds of silence to build a structure is normal, expected, and not a deduction on any scorecard. What costs you is skipping the pause and improvising as you speak, which produces overlapping branches almost every time.
Then run the same five moves in the same order on every case. Each move constrains the next, which is what stops you drifting back to a template when the prompt is unfamiliar.
Restate the objective as a decision or a number
Not "the client wants better profitability" but "find where 8 points of margin went, and say what would bring them back". If you cannot write the objective as a decision to make or a number to find, every branch after it is decoration.
Name the equation or the decision criteria
For a number, write the identity that produces it: profit equals revenue minus cost, revenue equals members times revenue per member, throughput equals the rate of the binding step. For a decision, write what would make the answer yes: payback inside three years, or a price the channel would accept. Most candidates skip this.
Cut into three or four branches that do not overlap
Three is usually right and four is the ceiling. Cut on one principle at a time so the branches are siblings rather than a mixed bag: by the equation, by customer segment, or by stage of a process, but not two at once. If two branches could hold the same fact, redraw now rather than eight minutes in.
Hang two or three drivers under each branch
This is where fit becomes visible. Drivers are the things you would actually go and look at: list price by tier, cost to serve per account, capacity added in the last two years. Three or four branches with two or three drivers each gives between six and twelve things to investigate, plenty for a thirty-minute case.
Name where you would start, and why
Pick one branch, say it is where you would begin, and give the reason in a clause. The reason earns the mark, because it shows commercial judgement rather than sequencing habit. "I would start with plan mix, because member numbers held flat while revenue per member fell" is a hypothesis, not a running order.
No framework was named, and the profit tree is visibly underneath it. That is the working relationship you want: the patterns supply the grammar, the prompt supplies the words.
How do you say a structure out loud?
A good structure delivered badly loses most of its value, because the interviewer marks what they can follow, not what is on your page. Four things do the work: signpost the shape first, keep each branch to one breath, state the hypothesis as a hypothesis, and hand back deliberately.
Signposting is the highest-return habit of the four. Saying "three parts, then where I would start" gives the interviewer somewhere to file your sentences, so a branch they half-missed does not cost you the whole structure. Here is the full delivery, on a case where a gym chain's revenue has fallen while member numbers held flat.
“I would like to look at this in three parts, and then tell you where I would start.”
“First, the fee itself: list price by tier, how much of it is being discounted, and whether old promotional rates are still running on renewals.”
“Second, plan mix: whether members have moved from annual to monthly, or from premium down to basic, and how much of the base has drifted.”
“Third, everything members buy beyond the fee: personal training, classes, guest passes.”
“My hypothesis is that this is mix rather than price. Member numbers have held flat while revenue per member has fallen, and that usually means people traded down rather than that everyone renegotiated.”
“So I would start with plan mix, and the first thing I would want is the split of members by tier for each of the two years. Does that sound like a sensible place to begin?”
That is about fifty seconds at a normal pace. It names three branches, attaches drivers without listing exhaustively, commits to a hypothesis with a reason, asks for one specific piece of data, and returns the floor. Every one is a scoreable move, and none needs a framework name.
The mid-build rule
Interviewers mark the structure you commit to, not a half-drawn one. That is an explicit guardrail on our rubric, and it changes how to handle a wrong turn. If you are three sentences into a cut and realise it overlaps, stop and redraw it in the open. Ploughing on to protect the appearance of confidence costs you the framing mark and everything downstream of it.
“Let me redraw that. I started splitting by region, but the question is about revenue per member, so region is a way to test the answer rather than a way into it.”
“Three parts instead: the fee itself, the plan mix, and what members spend on top.”
How do you practise structuring until it is reflex?
Structuring is the most trainable skill in the case, and it trains without a partner, which makes it good value for solo preparation. It is also the skill most people practise wrongly, by reading framework guides instead of producing structures. Reading builds recognition. Only building under time builds the reflex.
Reps under time
Take a one-line prompt, set ninety seconds, and produce a structure out loud. No case, no exhibits, no maths, just the opening. Fifteen of those fit in half an hour, where fifteen full cases would take a fortnight to arrange. The prompt need not be sophisticated: the five moves are identical whether the client sells insurance or ice cream. Our typed drills at /drills exist for this, but a notebook and a timer will take you a long way.
Self-marking against the framing anchors
After each rep, mark it against the three things framing rewards. Write the evidence before the score, because a number written first is a number you will find reasons for.
- Fit: does any part of this name something specific to this prompt, or would it survive a find-and-replace of the client?
- Separation: could the same fact land in two branches? If so, the cut was on more than one principle at once.
- Commitment: did I name a starting branch and give a reason, or did I list the branches and stop?
Score each 1 to 5 if you want the discipline of the real scale, where 3 means you met the first-round standard rather than that you were average. Fit and separation usually come first. Commitment is the one still at 2 after a fortnight, because naming a starting branch feels like a guess. It is supposed to.
Read your own tree back cold
Keep your structures and reread one the next morning, without the prompt in front of you. Two questions: can you reconstruct what the problem was from the structure alone, and would this fit any case in the book? The first should be yes and the second no. Fail the first and it was too abstract to guide anything. Pass the second and it is a template in your own handwriting.
This is also how you catch the drift back to memorised material. Nobody recites a framework on purpose: they get nervous and reach for whatever feels safest. Reps are what make the fitted version the safest thing available.
Then put it under live pressure
Silent practice has a ceiling: structuring on paper and structuring while somebody waits are different skills. Eventually you need a voice on the other end, a clock running, and a follow-up you did not expect. A three-minute taster at /try shows the gap, and a full case-only session runs thirty minutes, with the first one free.
Common questions
Do I need to memorise case interview frameworks at all?
Know the patterns, do not memorise the boxes. Six shapes cover most cases: profit tree, market entry, deal screen, pricing, growth, and capacity. Learn each as an equation or a decision criterion so it comes to hand automatically, then build the actual branches from the prompt in front of you. Reciting a framework verbatim scores worse than three branches you built yourself.
Is the profitability framework still worth learning?
Yes, more than any other, because the underlying identity is always true: profit equals revenue minus cost, and each side splits further. What is not worth learning is the standard four-box version. Cut the tree where the specific problem is interesting, so if volume rose while profit stayed flat, put the detail on price and cost to serve rather than splitting both sides evenly.
How many branches should a case interview structure have?
Three is usually right, four is the ceiling, and two is generally a sign you have not finished thinking. What matters more than the count is that branches are siblings cut on a single principle, and that each carries two or three concrete drivers underneath. Five or more is normally a checklist rather than a plan, and it becomes impossible to talk through cleanly.
Can an interviewer really tell that a framework was memorised?
Almost always, and quickly. The cadence smooths out because the words are being retrieved rather than composed, the nouns go generic with nothing from the prompt in them, and there is usually one branch that belongs to the framework rather than to the problem. That last branch is the giveaway, because the candidate cannot say why it matters here when asked.
What if I get halfway through my structure and realise the cut is wrong?
Redraw it out loud. Framing marks the structure you commit to, not a partially drawn one, so correcting yourself mid-build costs nothing and reads as judgement. Say what you started to do, why it does not fit the question, and what you are replacing it with. Continuing with a cut you know overlaps costs you framing and everything downstream of it.
How long should I take to build a structure in a case interview?
Sixty to ninety seconds, asked for explicitly and taken in silence. That is normal and expected, and no scorecard treats it as a deduction. What does cost you is narrating your own thinking process while you build, or skipping the pause and improvising the structure as you speak it, which reliably produces branches that overlap. Ask for the moment, take it, then talk.