Most candidates who "build an issue tree" draw a topic list. Revenue, costs, customers, competition. The interviewer is not scoring topics. They are scoring whether you can turn one ambiguous question into a small set of non-overlapping, complete, and testable sub-questions — and then bet on which branch is true.
That is the thesis: an issue tree is a logic tree of questions, not a menu of themes. It sits next to two cousins that look similar on a whiteboard and do different jobs — the driver tree (arithmetic identity) and the decision tree (choices under uncertainty). After this piece you will be able to name which tool you are using, run three tests on every branch before you speak, and present two fully worked trees with an explicit hypothesis.
What an issue tree is (and what it is not)
An issue tree starts with one root question the client needs answered and splits it into sub-questions that, taken together, settle the root. Each layer inherits the same discipline: if the answers to the children are known, the parent's answer is known.
That idea is not case-interview folklore. Barbara Minto codified it for consulting problem-solving in The Minto Pyramid Principle: Logic in Writing, Thinking and Problem Solving (1996), the edition that added a full problem-solving section on top of the earlier writing book. In a McKinsey alumni interview, she states the grouping rules that still govern the whiteboard: the point above must summarize those below; ideas in a group must be of the same kind and in logical order; and the group must be MECE — mutually exclusive and collectively exhaustive.
McKinsey's interviewing page describes the problem-solving interview as checking how you structure ambiguous challenges, identify important issues, work facts and data, and form a recommendation. The issue tree is the structure half of that sheet. It does not replace the math half (that is often a driver tree) or the go/no-go half (that is a decision tree).
| Tool | Root object | What branches are | What "solved" means |
|---|---|---|---|
| Issue tree | A question | Sub-questions / hypotheses | You know which story is true |
| Driver tree | A metric | Algebraic inputs | You can rebuild the number |
| Decision tree | A choice | Acts and chance outcomes | You pick a strategy under uncertainty |
I stop listening when a candidate opens a go/no-go prompt with company–customers–competitors, or opens a profit miss with "market trends" as a sibling of "price" and "volume." Wrong tool, or right tool with broken siblings.
Three tests every branch must pass
Before you present, run each first-layer branch through three tests. Fail any one and rewrite.
1. Same-kind test. Every sibling must answer the parent in the same logical currency. If the root is "Why did operating profit fall $12M?", every first-layer branch is a candidate cause (or a MECE cause category). Mixing "volume," "pricing strategy," and "should we exit wholesale" fails: two causes and one decision. Minto's rule that grouped ideas must be "logically the same" is this test in one sentence.
2. MECE test. No overlap, no gap. Ask two questions out loud. Overlap: could one true fact land in two branches? Gap: could the real answer sit outside every branch? "Revenue" and "pricing" as siblings fail overlap, because price already sits inside revenue. "Revenue" and "cost" with no path for one-time impairments fail gap if the P&L has them. Cap the top layer at three or four branches; more usually means you have not chosen a decomposition axis.
3. Settling test. A branch earns its keep only if one data pull, one exhibit, or a crisp yes/no can keep or kill it. "What about customers?" fails. "Did average selling price fall, holding mix constant?" passes. Minto's problem-solving language treats a useful "issue" as something you can resolve, not a topic you can discuss forever. In the room, that is what lets you prioritize: you name the branch you would test first and the number you would ask for.
Ned's rule. If you cannot say, in one sentence, which exhibit would kill a branch, do not put that branch on the board. The sheet rewards a falsifiable map, not a brainstorm.
How the hypothesis sits on the tree
The tree maps what could be true. The hypothesis names what you bet is true and which branch you open first. They are not the same object.
Say both out loud: "My working hypothesis is that the miss is volume in the wholesale channel; I will start there because wholesale is 70% of units, so a mid-single-digit volume drop can explain the whole dollar gap before I touch price or cost." Then walk the tree so the interviewer sees you have not closed your mind — you have ordered the work.
In a typical first round of eight candidates, my estimate is that two or three put a real hypothesis on the table in the first ninety seconds. The rest either never state one, or state one that no branch on their tree can falsify. The second group sounds decisive and scores poorly on structure.
Worked tree 1: Northline Apparel's $12M profit miss
Invented company, round numbers you can reproduce.
Northline Apparel sells mid-market outerwear through wholesale (department stores) and its own DTC e-commerce. Last year operating profit was $48M. This year it is $36M. Same SKU set. CEO asks why.
Root question: Why did Northline's operating profit fall $12M year over year?
Decomposition axis: Algebraic P&L first (profit = revenue − cost), then split revenue by price vs volume, and cost by fixed vs variable. That keeps siblings same-kind and MECE by construction.
Why did OP fall $12M?
├── Revenue down?
│ ├── Price (ASP) down?
│ └── Volume down?
│ ├── Wholesale units down?
│ └── DTC units down?
└── Costs up?
├── Variable cost per unit up?
└── Fixed cost up?
Hypothesis (marked): Volume in wholesale explains most of the $12M. Wholesale was 2.1M of 3.0M units last year; a 8% wholesale unit decline at last year's contribution margin of about $18/unit is roughly 2.1M × 0.08 × $18 ≈ $3.0M of contribution lost — not enough alone. So either the volume drop is larger, price moved with it, or cost also moved. The hypothesis is still the opening bet; the tree tells you what else must be true if the gap is $12M.
What I ask first: Wholesale units YoY, ASP by channel, and variable cost per unit. One exhibit that shows wholesale units −15% and ASP flat settles the volume branch as the primary story and points the next layer at why wholesale ordered less (sell-through, inventory, competitor share — a new issue tree under that leaf, not new siblings at the top).
Notice what this tree is not. It is not a driver tree yet: I have not written the full identity that rebuilds $12M from leaves. Once the volume branch survives, I switch tools and quantify. That handoff is intentional.
Worked tree 2: ClearBay Diagnostics' renewal drop
Different prompt, different axis. ClearBay sells software to mid-size clinical labs on annual licenses. Net renewal rate fell from 92% to 84% in one year. Product still ships. CEO asks why renewals fell.
Root question: Why did ClearBay's net renewal rate fall eight points?
Decomposition axis: Process / customer journey for renewals, not a P&L formula. Same-kind test: every branch is a stage where a renewing customer can be lost.
Why did renewals fall 8 pts?
├── Did fewer accounts reach renewal decision?
│ ├── Churned mid-term (contract exits)?
│ └── Acquired accounts never activated?
├── Of those who reached renewal, did more decline?
│ ├── Price / commercial terms rejected?
│ ├── Product / value perceived as weak?
│ └── Competitor displacement?
└── Did the renewing base mix shift?
├── Lost high-renewal segments?
└── Gained low-renewal segments?
Hypothesis (marked): Product / value among accounts that reached renewal explains most of the eight points — specifically, labs that adopted a competitor's LIS module and no longer need ClearBay's overlapping module. I open the "decline at renewal" branch first because an eight-point move is hard to explain with mix alone unless segment weights swung violently, and mid-term exits would already show in the churn dashboard the CEO would have named.
What I ask first: Of non-renewals, share citing competitor displacement vs price vs unused product; renewal rate by segment; mid-term exit rate YoY. If competitor displacement is 60% of non-renewals and non-renewals are ~16% of the base (100% − 84%), the competitor story carries most of the gap and the commercial and mix branches can wait.
Same three tests: siblings are all causes of renewal-rate decline; MECE along the journey; each leaf dies with one cut of the non-renewal reasons table.
How to build one in ninety seconds
- Write the root as one measurable question. Include the metric, direction, and magnitude when you have them. "Why did OP fall $12M YoY?" beats "profitability issues."
- Pick one decomposition axis. Algebraic if a formula exists. Process if the problem is a flow. Segmentation (geo, product, segment) if the unknown is where, not why. One axis per layer.
- Draw three or four first-layer branches. Run the three tests. Cut anything that fails.
- Mark the hypothesis and the first data request. Then present top-down: root, branches, bet, ask. A silent tree earns nothing on the sheet.
If the prompt is already a choice under uncertainty ("build, pilot, or wait"), do not force an issue tree. Use a decision tree. If the prompt is "explain this dollar miss" and you already know the identity, jump to a driver tree and prove the lever. Issue trees earn their keep when the question is still open.
For framing the first ninety seconds of the case itself, pair this with how to structure your thinking and clarifying questions. For picking a structure without naming a memorized framework, see the frameworks guide.
Practice this today
Take any profitability or churn prompt. Spend ninety seconds building only the first layer. Run the three tests out loud. State a hypothesis and one data request. Then rebuild the same prompt as a driver tree and notice where the tools diverge.
CoachNed's structure drills score the split and the prioritization. The case structure builder lets you draft a tree against a blank prompt before you take it live. Everything is open for seven days, no card; then $120 for a recruiting season or $49 a month. CoachNed is not affiliated with McKinsey, BCG, Bain, Deloitte, or any firm named above.
Frequently asked questions
What is an issue tree in a case interview?
It is a hierarchical breakdown of one root question into MECE, same-kind, testable sub-questions. Interviewers use it to see whether you can frame an ambiguous problem before data arrives. It is not a list of topics and not the same tool as a driver tree or a decision tree.
How is an issue tree different from a driver tree?
An issue tree settles which story is true. A driver tree rebuilds a metric from algebraic inputs so you can quantify the lever. Use the issue tree to frame; switch to the driver tree when you need the dollar proof. See the sibling driver tree guide for the arithmetic version.
How many branches should an issue tree have?
Three or four at the top layer is enough for almost every interview prompt. More usually means you have not chosen a single decomposition axis. Depth of two or three layers is plenty; deeper leaves wait until the parent branch survives contact with data.
Do I need to say MECE out loud?
No. Show it. Name the split cleanly, run a one-line overlap/gap check if you are unsure, and move on. Saying "this is MECE" without the tests is noise. Failing the tests with or without the acronym is what the sheet catches.
Should every case start with an issue tree?
No. Start with the tool that matches the prompt. Diagnosis of an unknown cause: issue tree. Quantification of a known identity: driver tree. Choice among acts under uncertainty: decision tree. Forcing one template onto every prompt is how candidates sound rehearsed instead of sharp.
Sources
- The Minto Pyramid Principle textbook (Barbara Minto) — primary source for the 1996 logic-in-problem-solving edition and pyramid/grouping rules behind issue trees. Checked 2026-09-24.
- Barbara Minto on MECE and the Pyramid Principle (McKinsey alumni) — Minto's own statement of MECE, summary-from-below, and same-kind grouping. Checked 2026-09-24.
- Interviewing at McKinsey — firm's public description of the problem-solving interview (structure, issues, data, recommendation). Checked 2026-09-24.
- The Pyramid Principle (Pearson / FT Publishing) — widely available writing/thinking edition of Minto's method. Checked 2026-09-24.