I stop listening when a candidate says "I would segment the customers" and then lists age bands. That is marketing taxonomy, not diagnosis. In most profitability and growth cases, the interviewer is waiting for a different move: cut the aggregate that looks fine until you split it.
Segmentation in a case is an analytical cut on revenue, cost, product, channel, or geography so you can see where the problem actually lives. The average is not the business. It is often two businesses canceling each other out.
After this piece you will pick a cut that can hide a problem, run the mix arithmetic that proves it, and refuse to solve a blended number that is lying to you.
Why the aggregate is usually the wrong patient
McKinsey's published problem-solving guidance puts disaggregation early: Charles Conn describes logic trees that take a problem apart into logical pieces before you prioritize which piece to attack. Bain's interviewing pages say the firm wants to see whether you think in a logical and structured way on realistic client problems, not whether you can recite a customer-segment list.
In the room that translates to a simple score: did you treat the headline KPI as a patient, or as a symptom of mixed parts?
| Headline the client gives you | What often hides underneath |
|---|---|
| "Revenue is flat" | One channel up, another down by the same dollars |
| "Margin held" | Price up, mix down into lower-margin SKUs |
| "Costs rose 4%" | One plant blew up while three plants improved |
| "Share is stable" | Gains in one region offset losses in another |
If you start solving the average, you will recommend an average fix. Average fixes fail in front of partners for the same reason they fail with clients.
This article is about that diagnostic cut. If your question is whether a customer split earns points only when segments behave differently on the commercial driver, use the sibling note on customer segmentation. Here the object of the cut is the P&L, not the persona deck.
Pick one axis that can hide the issue
A useful first cut is MECE by construction: one dimension, exhaustive buckets, no cross-cuts in the same layer. Mix geography and product in one list and you double-count.
| Cut | Use it when you suspect… | Metric to compare |
|---|---|---|
| Product / SKU family | Portfolio mix is shifting | Contribution margin, returns |
| Channel | Route-to-customer economics differ | Revenue, CAC, fulfillment cost |
| Geography | Local demand, cost, or competition | Growth, margin, share |
| Customer type (coarse) | One buyer class drives the swing | Retention, AOV, cost-to-serve |
| Cost bucket / process step | The pain is operational, not commercial | Unit cost, waste, labor hours |
Name the cut out loud, then name what would make a branch "guilty." "I would split revenue by channel because online and retail have different fulfillment costs; if online grew while retail fell, the blended margin would look stable while unit economics flipped." That sentence is the structure. The labels alone are not.
Aim for three to five branches. Two is often too coarse to localize. Six-plus is theater.
The arithmetic the sheet actually rewards: mix effects
Managerial accounting treats sales mix as a first-class driver of profit. When products (or channels, or regions) carry different contribution margins, a shift in proportions changes the blended margin even if total volume and list prices hold still. Standard CVP teaching notes make the point explicitly: if the mix shifts from high-contribution to low-contribution products, the average contribution-margin ratio falls and net operating income declines for a given sales dollar — a "profit problem" with no volume collapse in the headline.
FP&A practice calls the same idea price-volume-mix. FTI Consulting's PVM note (May 2024) walks a worked electronics portfolio where a 20.1% revenue increase still needed a mix bridge: phones declined while computers and tablets carried growth, and mix alone explained a material share of the dollar change. You do not need their full three-step formula in a case. You do need the intuition: volume, price, and mix are different stories.
In interview math, the smallest useful mix check is:
- Blended contribution = sum of (segment share × segment contribution rate).
- Hold total dollars or units fixed.
- Change only the shares.
- Read the delta as the mix effect.
If that delta is large, the problem is composition, not "the business." If it is small, stop segmenting and move to a driver inside the guilty branch.
Ned's rule. Never trust a blended margin until you have asked whether two segments moved in opposite directions by roughly the same amount. If the interviewer has not given you the cut, request it. Solving the average when an offsetting pair exists is how strong candidates still get a "no."
Worked example: the flat margin that is not flat
Prompt. Harborline Softgoods sells workwear through two channels: direct e-commerce and wholesale to industrial distributors. Year-over-year revenue rose from $100m to $104m. Contribution margin dollars rose from $28.0m to $28.1m. The CFO says "margin held, so the issue must be fixed cost or SG&A." Leadership wants a cost-cutting program.
A weak candidate accepts the frame and starts cutting HQ headcount. A stronger candidate asks for channel economics before touching SG&A.
| Channel | Year 1 revenue | Year 1 CM rate | Year 1 CM $ | Year 2 revenue | Year 2 CM rate | Year 2 CM $ |
|---|---|---|---|---|---|---|
| E-commerce | $40m | 35% | $14.0m | $55m | 34% | $18.7m |
| Wholesale | $60m | 23.3% | $14.0m | $49m | 19.2% | $9.4m |
| Total | $100m | 28.0% | $28.0m | $104m | 27.0% | $28.1m |
Check the aggregates the CFO used. Revenue up $4m. Contribution dollars essentially flat (+$0.1m). Blended CM rate down one point, easy to shrug off.
Now look at the offsetting pair.
- E-commerce added $15m of revenue and +$4.7m of contribution.
- Wholesale lost $11m of revenue and −$4.6m of contribution.
The two channels nearly cancel in contribution dollars. That is why the P&L looks calm. It is not calm.
Hold Year 2 total revenue at $104m and ask what contribution would have been if the Year 1 mix (40% e-commerce / 60% wholesale) had continued at each channel's Year 2 CM rate:
- Mix-held e-commerce revenue = 0.40 × $104m = $41.6m → CM = 0.34 × $41.6m = $14.1m
- Mix-held wholesale revenue = 0.60 × $104m = $62.4m → CM = 0.192 × $62.4m ≈ $12.0m
- Mix-held total CM ≈ $26.1m
Actual Year 2 CM is $28.1m. Relative to the mix-held baseline, the shift toward e-commerce helped by about $2.0m. The wholesale rate collapse (23.3% → 19.2%) is a separate, uglier story: on Year 2 wholesale revenue of $49m, that 4.1-point rate drop alone costs roughly $2.0m of contribution versus keeping the old wholesale rate.
So the "flat margin" hides two facts the cost-cutting program will miss:
- Growth is almost entirely e-commerce; wholesale is shrinking.
- Wholesale unit economics deteriorated while e-commerce stayed roughly healthy.
The recommendation flips. Do not start with a blanket SG&A cut. Investigate wholesale discounting, distributor mix, freight, and returns. Protect the e-commerce engine. The aggregate told you to cut; the cut told you where to look.
Reproduce the table with any two segments that move opposite ways. If their contribution deltas nearly cancel, you have found the interviewer's trap.
How to ask for the cut without stalling
McKinsey's interviewing guidance frames the problem-solving interview as a discussion of how you structure ambiguous challenges and deal with facts and data. Asking for a segment view is fair game when you explain why that view would change the diagnosis.
Three clarifying questions that earn their airtime:
- Do we have the KPI by product, channel, or region for the last two periods?
- Did any large share of the book grow while another shrank by a similar amount?
- Are contribution rates similar across those parts, or do they diverge by more than a few points?
If the interviewer says the data is not available, say what you would do with it and proceed with a proxy (for example, AOV and return rate by channel). Do not invent a twelve-branch tree while you wait.
Common failures I score down
- Customer-first habit. Defaulting to demographics on a cost or operations prompt.
- Label lists. "Retail, online, international" with no metric and no data request.
- Cross-cut MECE breaks. "SMB, enterprise, digital" in one layer.
- Solving the average after seeing the cut. You get the channel table and still recommend a company-wide price increase.
- Mix blindness. Calling a rate change a "volume problem" when shares moved and rates did not.
Practice this on a live profitability case
Take any flat-margin or flat-revenue prompt and force one cut before you touch SG&A or a generic price lever. Write the two-period table, compute the canceling contribution deltas, and state the recommendation in one sentence that names the guilty segment.
A five-minute first rep is open at /start. For a full voice case with a debrief, use /interview. Math under time sits in /drills/math and the free /drills/quick-math set. 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
Is segmentation the same as a profitability framework?
No. Profitability starts with the identity revenue minus cost. Segmentation is how you disaggregate a branch of that identity when the branch is mixed. Use the formula first; cut when the formula's pieces look too average to trust.
When should I segment costs instead of revenue?
McKinsey's published problem-solving guidance puts disaggregation early: Charles Conn describes logic trees that take a problem apart into logical pieces before you prioritize which piece to attack. Bain's interviewing pages say the firm wants to see whether you think in a logical and structured way on realistic client problems, not whether you can recite a customer-segment list.
When the client complaint is unit cost, waste, overtime, or plant variance, and revenue looks healthy. Split by plant, SKU, process step, or shift. The mix logic is identical: one bucket can explode while the blended cost looks mild.?
A useful first cut is MECE by construction: one dimension, exhaustive buckets, no cross-cuts in the same layer. Mix geography and product in one list and you double-count.
How many segments should I propose in a case interview?
Three to five on one axis. Enough to localize, few enough to analyze. Add a small "other" bucket if you need exhaustiveness.
Do I need price-volume-mix formulas memorized?
No. Know that mix can move profit with volume and price held conceptually fixed, and be able to rebuild a blended rate from shares and segment rates. Formal PVM bridges are for FP&A decks; the interview rewards the insight and the clean arithmetic.
How is this different from customer segmentation?
Customer segmentation asks whether buyer groups behave differently on the driver that decides the offer. This guide is about cutting any commercial or operational aggregate to find where the problem lives. Same word, different job.
Sources
- How to master the seven-step problem-solving process (McKinsey) — Conn on logic trees and disaggregating problems before prioritization. Checked 2026-09-24.
- Interviewing at McKinsey — Official framing of the problem-solving interview as structuring ambiguous challenges and working with data. Checked 2026-09-24.
- Interviewing | Bain & Company — Bain's public statement that case interviews test logical, structured thinking on realistic client problems. Checked 2026-09-24.
- A Quantifiable Approach To Price Volume Mix Analysis (FTI Consulting) — Worked PVM bridge showing mix as a distinct revenue driver (May 2024). Checked 2026-09-24.
- Cost-Volume-Profit Relationships (UPR teaching notes, Ch. 6) — Sales mix as a driver of the average contribution-margin ratio and break-even. Checked 2026-09-24.