Short answer
An issue tree breaks a problem into smaller questions, level by level, until each branch can be answered with data. Use a diagnostic (why) tree to find causes, a solution (how) tree to generate options, and a hypothesis tree to test a likely answer fast. A good tree is MECE, uses one logic per level, and ends in branches you can test.
Key takeaways
- An issue tree splits one question into sub-questions until each can be answered with a specific piece of data.
- Diagnostic trees ask why, solution trees ask how, and hypothesis trees start from a likely answer and list what must be true.
- McKinsey's seven-step problem-solving process puts logic trees at step two and recommends drawing two or three different cuts.
- Hypothesis-driven work means stating a likely answer early, testing it, and revising it as soon as the data disagree.
- Trees are for analysis; the Minto pyramid is for presenting the answer once the analysis is done.
What is an issue tree?
An issue tree is a diagram that breaks one question into smaller questions, level by level, until every branch can be answered with a specific piece of data. The top is the client's problem. Each level below splits the level above into parts that do not overlap and together cover everything. Consultants use issue trees to plan analysis; candidates use them to structure a case.
The method is standard consulting practice. In a September 2019 McKinsey podcast, former partner Charles Conn and senior partner Hugo Sarrazin describe a seven-step problem-solving process in which step two is disaggregating the problem with logic trees. Conn and Robert McLean later set out the method in the book Bulletproof Problem Solving (Wiley, 2019, reported).
Diagnostic, solution and hypothesis trees: which to use when
There are three common kinds of issue tree, and each answers a different question. Pick the one that matches the client's question.
| Tree | Question it answers | Top of the tree | Use it when |
|---|---|---|---|
| Diagnostic (why) | Why did this happen? | The problem, quantified | Profit fell, costs rose, share dropped |
| Solution (how) | How could we fix or achieve this? | The goal | The cause is known and the client wants options |
| Hypothesis | Is this likely answer true? | A specific claim | You have a strong first guess and limited time |
A driver tree is a diagnostic tree in which every branch is a number, so branches add or multiply up to the top. It is the most useful form in profitability cases.
How to build an issue tree in five steps
Building a tree takes five steps, and the first one matters most:
- Write the question precisely. Include the metric, the size and the time frame: "Why did cost per passenger-km rise 15% in two years?"
- Choose the first cut. Prefer a mathematical split (cost divided by volume) or the client's own units (routes, stores, products).
- Split each branch again with one logic per level, for example by cost type, then by cause.
- Stop when a branch maps to data you could ask for in the interview.
- Mark where you would start, usually the branch you expect to be largest.
Conn recommends drawing two or three different cuts of the same problem, because each cut reveals something the others hide. In an interview, sketch two first cuts quickly and present the one that leads to data fastest.
What does hypothesis-driven mean in a case interview?
Hypothesis-driven means stating a likely answer early and using the tree to test it before gathering data on every branch. McKinsey teams aim for a quick "one-day answer" and then iterate (official, 2019 podcast). Firms ask candidates for the same habit: Deloitte asks for an initial hypothesis after the structure, and Strategy& Netherlands asks candidates to test hypotheses with numbers.
The skill lies in revising. A hypothesis is useful only if you drop it as soon as the data disagree, and say so out loud. Our case study interview guide shows where this fits in the case flow.
Worked example: one problem, three trees
Prompt (fictional): A Dutch regional bus operator's cost per passenger-km rose from €0.40 to €0.46 in two years, a 15% increase. Passenger-km stayed at 300M, so total cost rose from €120M to €138M, an €18M gap. The board asks why and how to cut cost.
(a) Diagnostic tree: why did cost per passenger-km rise?
- Cost per passenger-km up €0.06
- Passenger-km (flat at 300M, so not the driver)
- Passengers
- Average trip length
- Total cost up €18M
- Driver labor: regular hours, overtime, wage rates
- Fleet: depreciation, energy, maintenance
- Overhead: depots, planning, IT
- Passenger-km (flat at 300M, so not the driver)
(b) Solution tree: how could the operator cut cost per passenger-km?
- Lower cost per passenger-km
- Raise passenger-km: ridership on weak routes, better timetables
- Lower labor cost: hire to cut overtime, smarter rostering
- Lower fleet cost: off-peak charging, maintenance contracts, longer asset lives
- Lower overhead: merge depots, share planning tools
(c) Hypothesis tree: "Rising driver overtime explains most of the €18M gap."
| What must be true | Data test | Result (fictional) |
|---|---|---|
| Overtime cost grew by more than half the gap | Overtime cost 2024 versus 2026 above €9M | Grew €6M: fails |
| Other cost lines grew less | Fleet, maintenance and other cost changes | Fleet +€8M (depreciation +€10M, energy -€2M), maintenance +€2M, other +€2M |
| The cause is structural | Driver vacancies and absence rates | Vacancies up, so overtime will persist |
The hypothesis fails: overtime explains €6M of €18M, or 33%, while fleet costs explain €8M, or 44%. Check: 6 + 8 + 2 + 2 = 18. The candidate should say so, move the hypothesis to fleet electrification ("new electric buses raise depreciation faster than they cut energy cost"), and open that branch next. Overtime still deserves a recommendation, because the €6M is real and its cause persists.
Five tests of a good tree
An interviewer can judge a tree in about 30 seconds with five checks:
| Test | Question to ask | Failure example |
|---|---|---|
| Mutually exclusive | Could one euro of cost sit in two branches? | "Labor" next to "overtime" at the same level |
| Collectively exhaustive | Do the branches add up to the whole? | Cost tree with no overhead branch |
| One logic per level | Is each level split the same way? | Mixing cost types and regions in one level |
| Testable | Does each leaf map to data you could request? | A leaf called "efficiency issues" |
| Prioritized | Is the likely largest branch marked first? | Ten equal branches with no starting point |
Our MECE guide covers the first two tests in depth.
From issue tree to pyramid: analysis first, communication second
Issue trees and the Minto pyramid are two halves of the same job. The tree runs from question to data: it plans the analysis. The pyramid runs from answer to support: it presents what the analysis found. In the bus case, the tree found that fleet costs explain 44% of the gap. The pyramid opens with that answer, then groups the evidence beneath it.
Candidates often present their tree as the conclusion. Present the answer first instead, as our guide to the Minto Pyramid Principle explains, and keep the tree as the map behind it.
A copy-ready issue tree template
Copy this three-level template and fill it in before each practice case:
- Question: [metric] [changed by how much] [over what period], or [goal] by [date]
- Driver 1: [largest expected branch]
- Sub-driver 1a: [data you would request]
- Sub-driver 1b: [data you would request]
- Driver 2:
- Sub-driver 2a:
- Sub-driver 2b:
- Driver 3:
- Sub-driver 3a:
- Sub-driver 3b:
- Hypothesis: [one sentence] / Start with: [driver]
- Driver 1: [largest expected branch]
For which first cuts suit which case types, see our case interview frameworks guide, and practice on full cases in the complete case interview guide. Cracking the Case walks through three full candidate cases with the structures drawn out.
Frequently asked questions
What is the difference between an issue tree and a decision tree?
An issue tree breaks a problem into its components or causes. A decision tree maps choices and uncertain outcomes with probabilities to compare expected values. Case interviews mostly use issue trees.
Is an issue tree the same as a case framework?
A case framework is usually the first one or two levels of an issue tree, drawn for the interviewer. The full tree continues down to the data you would need.
How many levels should an issue tree have?
Three levels are enough in an interview: the question, three to five drivers, and two to four sub-drivers each. Go deeper only on the branch the data points to.
What is a driver tree?
A driver tree is a diagnostic tree in which every branch is a number linked by arithmetic, such as profit equals revenue minus costs. It lets you quantify which branch explains a change.
Who invented the issue tree?
No single inventor is credited. Logic trees became standard practice at McKinsey and other firms, and former McKinsey partner Charles Conn describes them as step two of the firm's seven-step process.