PartnerRoundBy the end you can
- Spot a split that is not MECE and fix the gap or overlap in seconds
- Pick the right first split (mathematical, process or segment) for any prompt
- Build a two-level tree in the client's own language and present it in under a minute
- Stress-test your own tree with five quick checks before you speak
On this page
01What MECE means
MECE stands for mutually exclusive, collectively exhaustive. Mutually exclusive: no item can sit in two branches. Collectively exhaustive: every possible answer has a branch to sit in.
An overlap wastes time and double counts. A gap is worse: the answer may sit exactly where you never looked. Interviewers do not expect perfection, but they notice both within seconds.
Check yourself
A gym chain's membership is falling. Which first split is MECE?
02Three ways to split a problem
Almost every good first split is one of three kinds. Know all three and you will not freeze on an unfamiliar prompt.
1. Mathematical
An equation is MECE by construction: nothing can sit outside it.
If the question is about a number, make the first level an equation. Sales = market size × share. Revenue = customers × purchases per customer × price. Members = start + joiners - leavers.
2. Process or value chain
- 01OrderPlaced and released to the warehouse
- 02Pick and packItems collected, checked, bagged
- 03Load and routeVans loaded, routes planned
- 04DeliverDrive, find the door, hand over
Use a process split when the problem lives in how work flows: delays, errors, cost to serve, customers dropping out. Follow the order or the customer from start to finish and nothing gets skipped.
3. Segment
Split by who, what or where: customer type, product line, geography, channel. Use it when results differ across the business, which is nearly always. Segments are the classic second level under a mathematical first level: revenue, then revenue by product.
03Hypothesis-driven vs issue-driven trees
An issue-driven tree breaks the question into sub-questions with no view on the answer. A hypothesis-driven tree starts from a likely answer and lists what must be true for it to hold.
In an interview you need both. Lay out an issue-driven tree so the interviewer sees you have covered the problem, then use a hypothesis to choose where to dig first. If the data kills the hypothesis, say so and move to the next branch: that is the method working, not failing.
04How deep to go, and making it fit
Say two levels out loud and keep a third on paper where you have a reason. Two to four branches at each node gives you six to twelve ends: enough to show breadth, few enough to say in a minute.
Stop splitting when a branch points to data you could ask for or an analysis you could run. "Customer satisfaction" is not an end. "Repeat purchase rate by customer group over the last three years" is.
Then make it fit. A generic framework recited word for word is the most common reason a structure scores average. Keep the skeleton, but name every branch in the client's language and say why it matters for this business.
Weak
I'd look at revenue and costs. Revenue is price times volume, and costs split into fixed and variable. Then I'd look at the market, the competitors, the customers and the company itself.
Strong
Profit is membership fees plus extra revenue, minus costs. On memberships I'd split the count into new joiners and cancellations, because gyms live or die on churn, and check the fee per member after discounts. Extra revenue is personal training and classes. Most gym costs are fixed: rent, staff, equipment leases, so I'd check whether new sites opened that have not filled yet. And I'd look at all of it site by site, because 120 urban gyms will vary a lot.
Why: Both use the same profit skeleton. The strong answer names gym drivers (joiners, cancellations, personal training, new sites still filling up) and gives a reason for each branch. The interviewer hears judgement about this business, not a memorised list.
05Present your tree in under a minute
Ask for a minute to think, draw the tree in silence, then turn your page so the interviewer can see it. Number the branches. Say the top level in one breath before going into any branch.
- 01GoalRestate the objective in one line
- 02Top level"I'd look at three areas"
- 03Each branchTwo or three items, with why
- 04PriorityWhere you would start, and why
- 05First askThe data you want first
That takes about 40 seconds to say. If your delivery runs past 90 seconds, you have too many leaves: name the two most important items under each branch and keep the rest on the page for later.
06Test your tree for gaps and overlaps
Before you speak, spend ten seconds on five tests. They catch most of the errors interviewers mark down.
- Sum test. If the root is a number, do the first-level branches add or multiply up to it? If not, there is a gap.
- Overlap test. Take a concrete event, such as a rival cutting prices 10%, and count how many branches it lands in. More than one means an overlap.
- Gap test. Name the three likeliest answers to the case. Each needs an obvious home. If one has nowhere to go, add a branch.
- Level test. Branches at the same level should be the same kind of thing. "Revenue, costs, the new CEO" mixes drivers with a cause.
- So-what test. Would finding something in this branch change the recommendation? If not, cut it.
Worked example
Worked fix: a coffee chain's falling sales
Prompt. A chain of 200 coffee shops has seen sales fall from $240m to $216m in a year, down 10%. Why?
First draft. Competition, prices, product quality, marketing, store locations.
Run the tests. A rival opening next door lands in both competition and store locations: overlap. Prices are something we set, competition is something done to us: mixed levels. And fewer office workers after a shift to hybrid working has nowhere to go: gap.
Rebuild it mathematically. Sales = stores × transactions per store × average ticket. Last year: 200 × 200,000 × $6.00 = $240m. This year: 200 × 180,000 × $6.00 = $216m. Stores and ticket are flat, so the whole fall is transactions per store, down 10%.
Segment that branch. The 100 stores in office districts fell from 220,000 transactions each to 180,000, down 18%. The 100 suburban stores held at 180,000. Every lost transaction is in office districts.
The likely root cause, hybrid working emptying office districts, was the gap in the first draft. The mathematical tree found it in two splits.
Check yourself
Your revenue tree's first level reads: new customers, existing customers, online customers. What is wrong?
Cheat sheet
The lines worth rereading the night before.
- 1MECE: nothing sits in two branches, and every possible answer has a branch.
- 2Three splits: mathematical (an equation), process (the steps in order), segment (who, what, where).
- 3If the question is a number, make the first level an equation. It cannot have a gap.
- 4Structure issue-driven, then prioritise with a hypothesis: "I'd start with X, because Y."
- 5Two levels out loud, two to four branches each. Stop when a leaf points to data you could ask for.
- 6Use the client's language, never a framework's name.
- 7Present in under a minute: goal, top level, each branch, where you'd start, first data request.
- 8Before you speak, run the sum, overlap, gap, level and so-what tests.
Now drill it
Reading builds the model; drills make it automatic. Start with the first topic, easiest drills first.
- Profitability7 drills · Find out why profit moved by splitting it cleanly, segmenting early and asking the question that localises the problem.
- Market entry7 drills · Decide whether a client should enter a market, and how, with a structure that ends in a go or no-go.
- Growth ideas7 drills · Brainstorm ways to grow revenue in clear buckets, then say which ideas you would pursue first.
Read the full guide
The same ground on the blog, with more worked examples and practice on the page.
Questions candidates ask
What does MECE stand for?
Mutually exclusive, collectively exhaustive. The idea is associated with Barbara Minto, who developed the Pyramid Principle at McKinsey. In a case it means your branches do not overlap and, together, cover every possible answer.
Is an issue tree the same as a case framework?
A framework is a ready-made issue tree for a common case type, such as profitability or market entry. The tree is the underlying skill. Interviewers reward trees built for the client in front of them, so treat frameworks as starting points and adapt them every time.
Does my structure have to be perfectly MECE?
Your first level should be clean, and any mathematical split should be exact. Deeper qualitative levels can be close enough, as long as nothing important is missing and nothing is counted twice. Relevance to the client matters more than theoretical purity.
How long should I take to build an issue tree in a case interview?
Asking for a minute or two to structure is normal and expected. Aim to draw the tree in 60 to 90 seconds and present it in under a minute. If you need much longer, the tree is probably too deep.
What is the difference between an issue tree and a hypothesis tree?
An issue tree splits the question into everything that could explain it. A hypothesis tree starts from a likely answer and lists what must be true for it to hold. Strong candidates usually present an issue tree first, then use a hypothesis to choose where to start.