The room goes quiet after the prompt, and nothing slides across the table. That silence is the test. Most candidates treat an unstructured case as permission to brainstorm themes; the scoring sheet rewards something narrower: whether you can invent the next exhibit yourself and still close.
After this page you should be able to turn a messy one-line brief into a three-branch workplan, two prioritized data asks, a bound if the interviewer has nothing, and a synthesis that does not wait for a chart that is not coming.
What "unstructured" actually means
Firms do not publish a product called the unstructured case. What shows up in partner rounds, boutique finals, and some candidate-led offices is a prompt with no prepared exhibit pack. You may earn a number later if you ask well. You may never see one.
That is different from three formats candidates often confuse it with:
| Format | Who drives | What you get | What fails |
|---|---|---|---|
| Interviewer-led (classic McKinsey style) | Interviewer sequences questions | Exhibits and math prompts arrive on a schedule | Freezing when they ask the next discrete question |
| Candidate-led with exhibits | You choose the next analysis | Charts appear when you ask the right thing | Boiling the ocean or ignoring the pack |
| Unstructured / thin-prompt | You invent the analyses | Often nothing until you earn it | Waiting, PESTLE dumps, TED talks |
McKinsey's interviewing page describes a problem-solving interview around a client scenario and posts practice cases with sequenced questions and exhibits. BCG's case preparation guidance stresses structuring the approach, asking thoughtful questions, and reasoning when there is not a single right answer. Bain's case prep page tells candidates to clarify the objective, structure thinking, and adapt as the conversation evolves. None of those pages uses the word "unstructured." The label is candidate folklore for "thin prompt, thin data." Treat it as a skill, not a named product.
What the sheet rewards when the pack never arrives
When I score a thin-prompt case, I am not grading industry trivia. I am listening for four beats, usually in this order:
- Decision and metric. Restate the decision. If success is ambiguous, pick one primary metric and name the trade-offs you are parking.
- Analyses, not themes. Branches must be work you could staff on Monday, not "political / economic / social."
- Hypothesis that cuts the tree. Say where you think the answer lives so you do not boil the ocean.
- Prioritized asks, then a close. Ask for the one or two facts that would change the recommendation. If you get shrugs, bound the answer and still recommend.
The leak is almost always beat two. Candidates who practiced only exhibit-heavy cases wait for the chart. Candidates who over-memorized frameworks dump MECE wallpaper. Neither looks like an associate on day two.
Ned's rule. If a branch cannot become a two-week workstream with an owner and a data source, it is not a branch. Themes decorate slides; analyses staff teams.
A worked example you can reproduce
Prompt. "HarborMart, a 40-store regional grocery chain, wants same-day delivery without wrecking store margins. How would you think about it?"
No pack. No market share deck. No cost curve.
Clarify (45 seconds). Decision: launch, pilot, or kill same-day. Primary metric: contribution margin per order after delivery cost, not GMV. Time: board wants a view in 90 days. Assume they confirm.
Tree (analyses you invent):
| Branch | Question the branch answers | First data ask |
|---|---|---|
| Demand density | Which zip codes have enough orders per hour to fill a driver? | Orders/hour by zip for the last 12 weeks |
| Unit economics | Does contribution stay positive after last-mile? | Basket margin, pick cost, delivery cost per drop |
| Store ops | Does picking steal labor from checkout and spoil in-store NPS? | Pick minutes per order and overtime at peak |
| Competitive necessity | Is same-day table stakes or a vanity race? | Share of local peers already offering under 2 hours |
Hypothesis. The answer lives in density and unit economics, not branding. If fewer than three zips clear a break-even order rate, pilot those zips only; do not network-wide.
Math you invent when they give you nothing. Suppose you bound: average basket contribution before delivery is $18; pick cost $3; last-mile $9 if you hit 4 drops per hour, $14 if you hit 2. Contribution after delivery is $18 - $3 - $9 = $6 at 4 drops/hour, and $18 - $3 - $14 = $1 at 2 drops/hour. At one drop/hour you are underwater. That bound is the exhibit you invent: a density threshold of roughly 3–4 drops per driver-hour before HarborMart should expand the map.
Close (60 seconds). Pilot same-day in the top three zips by order density only if those zips clear ~3 drops per driver-hour at current basket mix. Kill network-wide launch. Next two weeks: pull order heatmaps and time three peak shifts of pick-and-pack. Risk: competitors already cover those zips and you are buying volume you cannot keep.
You never saw a chart. You still did a case.
The five-minute product: tree plus workplan
Think of the unstructured answer as two slides, not a speech.
Slide 1 is the tree. Three or four mutually exclusive analyses that map to the decision metric. Congestion pricing is a useful mental model: success might be net revenue, traffic speed, equity, or downtown retail. Ask which one matters. For a real-world sense of how messy "policy plus money" gets, Transport for London's FOI breakdown shows Congestion Charge gross revenue of £286m in 2023/24 (£215m charge income plus £71m enforcement), which is exactly the kind of split an unstructured case would force you to invent before anyone hands you a table.
Slide 2 is the workplan. For each branch, name the data, the owner analogue ("ops / finance / research"), and the kill criterion. Interviewers who will not give you exhibits still give points when you show you know how the work would run.
A bad tree sounds like: "I'd look at customers, competitors, company, and context." A good tree sounds like the HarborMart table above.
What not to do (and why it costs you points)
- Interview the interviewer for ten minutes. Clarifying is fine; fishing for the secret answer is not. Cap yourself at two or three clarifying questions, then structure.
- Dump PESTLE or a named framework. BCG's own brochure language has long valued analytical practicality over "systematic use of standard management concepts." Naming Porter does not save a thin-prompt case.
- Wait for the exhibit. Silence after your structure is your cue to propose the first analysis and ask for the fact that would change it.
- Give a TED talk. Insight without a decision is entertainment. Close with go / no-go / pilot and a next test.
People who train only on interviewer-led McKinsey-style cases are especially brittle here: they are used to the next question arriving. Candidate-led practice is the gym for inventing the next move yourself.
How I score a thin-prompt minute by minute
In a typical 25-minute partner case with no pack, the clock usually breaks like this. First two minutes: clarify and lock the metric. Next three: structure and state a hypothesis. Roughly twelve minutes: drive one or two branches with asks, bounds, and light math. Last three: synthesis and risks. The remaining time is theirs for follow-ups.
If you are still clarifying at minute eight, you have already lost the structure points. If you never close because you are "waiting for more data," you have lost the judgment points. I would rather hear a wrong-but-clear pilot recommendation with one named risk than a polished list of things you would research forever.
A practical self-check after a practice rep: mute the recording and write what decision you made, what metric you used, and what you would staff in week one. If any of those three is blank, the rep failed even if your voice sounded confident.
Practice this today
Tonight, close the tab after reading one headline. Set a 12-minute timer. Voice only. Deliver: metric, three analysis branches, two data asks, one bound if you get no numbers, a 45-second recommendation. Record it. Do not Google during the twelve minutes.
Useful prompts: a regulator floats a fee cap; a university pauses a program; a city cuts night bus frequency; a plant cannot staff third shift. After the timer, ask only one question of yourself: could each branch become a two-week workstream?
Then stay bilingual. Sit one full structured case so exhibit muscle does not atrophy. On CoachNed, start with a five-minute typed rep at /start, or the live voice case at /interview. Drill inventing branches below, then prioritization at /drills/synthesis. Everything is open for seven days, no card; then $120 for a recruiting season or $49 a month.
Frequently asked questions
Is an unstructured case the same as a written case?
No. Written cases usually hand you too much data in a packet and score how you select and synthesize. Unstructured live cases hand you too little and score how you invent the workplan.
Will the interviewer eventually give me numbers?
Sometimes, if your ask is specific enough to deserve an answer. Treat any number as a reward for a sharp tree, not a right you are owed. Plan to close with bounds either way.
Are unstructured cases only at boutiques?
No. Thin prompts show up in partner finals across firms and in offices that run more candidate-led discussions. Confirm the format with your recruiter; do not assume every round looks like the firm's online practice case.
How is this different from a brain teaser?
Brain teasers are puzzles with a clever trick. Unstructured cases are business decisions with missing inputs. If the prompt has a client and a choice, build a workplan, not a riddle answer.
Should I practice candidate-led or interviewer-led cases?
Both. Interviewer-led builds exhibit and math speed. Candidate-led builds the invent-the-next-move muscle unstructured rounds demand. If you only do one style, you will leak points in the other.
Sources
- Interviewing at McKinsey — firm description of the problem-solving interview and practice cases. Checked 2026-09-24.
- BCG case interview preparation — official guidance on structuring, questioning, and reasoning without a single right answer. Checked 2026-09-24.
- Preparing for the case interview | Bain — Bain tips on clarifying objectives, structuring, and adapting in the conversation. Checked 2026-09-24.
- BCG recruiting interview preparation brochure (Nov 2018) — durable firm language on practical analysis over rote frameworks. Checked 2026-09-24.
- TfL FOI: Congestion Charge and ULEZ revenue — primary figures for charge vs enforcement split (£286m Congestion Charge gross revenue in 2023/24). Checked 2026-09-24.