In a first-round case, I hear a clean customer tree about twice an hour. Age bands. SMB versus enterprise. Urban versus rural. The interviewer nods, then asks one question: so what would you do differently for each group? Half the time the answer is the same pitch, the same price, the same channel. That tree just cost you the points it was meant to earn.
A segmentation is only worth scoring if the segments behave differently on the driver that matters for the decision on the table. Split that changes willingness to pay, cost-to-serve, churn, or adoption. Split that does not is a slide, not a structure.
After this piece you will be able to reject a demographic cut in under ten seconds, name a needs-based alternative, and prove the flip with a small table of economics.
What Wendell Smith actually argued in 1956
Segmentation is not a case-interview invention. In July 1956, Wendell R. Smith published "Product Differentiation and Market Segmentation as Alternative Marketing Strategies" in the Journal of Marketing. Writing against the old assumption of homogeneous demand, he treated segmentation as a deliberate strategy: recognize that demand is heterogeneous, then align supply against distinct demand groups rather than forcing one product at everyone through advertising alone.
Two implications travel into the interview room.
First, segmentation is a choice about how you compete, not a taxonomy exercise. Smith framed it as an alternative to pure product differentiation. If your cut does not change the offer, the price, or who you chase, you have not segmented in his sense.
Second, the cut has to map to real differences in demand. Labels that look neat on a whiteboard but collapse into the same economics fail the test he set in 1956.
I do not expect you to cite the article in the room. I do expect your structure to survive the same question Smith was answering: why treat these buyers as different businesses?
Needs-based cuts beat demographic cuts for prediction
Demographic and firmographic variables are easy to observe. Age. Income. Headcount. Square footage. They are also weak predictors of purchase behavior when the job the customer is hiring the product for differs inside the same demographic cell.
That is not folklore. In 1968, Russell I. Haley argued in the Journal of Marketing that most segmentation techniques rely on descriptive factors about purchasers and are poor predictors of future buying; he proposed benefit segmentation — grouping first on the benefits people seek, which he treated as causal. Daniel Yankelovich made the parallel case for nondemographic criteria in HBR in 1964, then returned with David Meer in "Rediscovering Market Segmentation" (2006) to argue that psychographic casting for ads had drifted away from the original purpose: finding customers whose needs are unmet or whose behavior can change.
| Cut type | What it captures | Typical interview failure |
|---|---|---|
| Demographic / firmographic | Who the buyer is on paper | Same offer for both segments |
| Geographic | Where regulation, density, or logistics differ | Fifty regions with no thesis |
| Behavioral (usage, channel) | What people already do | Circular if defined by the outcome you want |
| Needs / benefits sought | Why they buy | Invented personas with no numbers |
Demographics are fine as a proxy when they correlate tightly with the need (clinician count and implementation cost often do). They fail when they do not. The interviewer is listening for whether you know the difference.
What the scoring sheet actually rewards
When segmentation is the right tool, the sheet is not grading your vocabulary. It is grading three moves:
- Name the commercial driver the decision hinges on (price power, cost-to-serve, retention, attach rate).
- Cut on an input that should move that driver, not on the outcome itself ("high-value customers" is circular).
- Show divergent economics or behavior, then change the recommendation.
MECE trees that produce identical actions for every branch score like a blank page. Overlapping cuts ("SMB / enterprise / digital natives") score worse: digital natives sit in both size buckets. If you need a refresher on that trap, see the MECE principle note on this site.
Failure mode: a live demographic cut that looks professional
Prompt. Ridgewell sells commercial HVAC filters and service contracts to building operators. Revenue growth stalled at 2%. The VP of sales wants a "segmented go-to-market." The deck on the table splits accounts by building size: under 100k sq ft, 100–400k, 400k+.
| Size band | Accounts | Avg contract | Gross margin | Churn |
|---|---|---|---|---|
| Under 100k sq ft | 1,800 | $4,200 | 41% | 18% |
| 100–400k | 620 | $18,000 | 44% | 11% |
| 400k+ | 140 | $62,000 | 46% | 9% |
A candidate who stops here recommends "focus on large buildings" — higher ACV, better margin, lower churn. It sounds adult. It is also wrong for the driver that matters.
The case exhibits then show utilization and complaint data. Small medical outpatient buildings reorder quarterly because filter specs are regulated and downtime is expensive. Large Class A offices stretch change-outs and hammer price in RFPs. The size cut mixed two needs inside each band: compliance-driven replacement versus cost-minimizing deferral.
Size correlated with ACV. It did not segment on the retention and attach-rate driver. That is the failure mode I see most often: a firmographic cut that produces a prettier average, not a different business.
Worked example: re-segmenting flips the recommendation
Prompt. Northline sells specialty adhesives into manufacturing. Book of 2,100 accounts. Growth is flat. Leadership's current grid is headcount: under 50 employees, 50–250, 250+. They want to staff an "enterprise hunter" pod because the 250+ band shows the highest ACV.
Demographic book:
| Headcount | Accounts | Avg ACV | Gross margin | Annual churn | Cost-to-serve |
|---|---|---|---|---|---|
| Under 50 | 1,240 | $6,500 | 52% | 22% | $900 |
| 50–250 | 680 | $24,000 | 48% | 14% | $3,200 |
| 250+ | 180 | $95,000 | 41% | 12% | $28,000 |
Contribution after cost-to-serve (rough):
- Under 50: (6,500 × 0.52) − 900 = $2,480
- 50–250: (24,000 × 0.48) − 3,200 = $8,320
- 250+: (95,000 × 0.41) − 28,000 = $10,950
On this cut, enterprise still looks best. The candidate who recommends the hunter pod is following the slide.
Now re-cut on application need — structural bonding (safety-critical joints in auto and aerospace suppliers), packaging adhesives (high volume, spec-light), and craft/DIY retail fill. Same accounts, different axis. You can say you would get application codes from CRM and order lines.
| Need segment | Accounts | Avg ACV | Gross margin | Churn | Cost-to-serve | Notes |
|---|---|---|---|---|---|---|
| Structural | 410 | $38,000 | 58% | 6% | $4,500 | Spec lock-in; co-developed with plant engineers |
| Packaging | 1,120 | $14,000 | 36% | 19% | $2,100 | Spot buys; three competitors in every RFP |
| Craft / DIY fill | 570 | $4,800 | 55% | 24% | $700 | Distributor-led; little direct sales lift |
Contribution after cost-to-serve:
- Structural: (38,000 × 0.58) − 4,500 = $17,540
- Packaging: (14,000 × 0.36) − 2,100 = $2,940
- Craft: (4,800 × 0.55) − 700 = $1,940
Cross-check against headcount: of the 180 "enterprise" logos, 95 are packaging converters with big plants and thin margins. Of the 410 structural accounts, 260 sit in the 50–250 band. The demographic "enterprise" bucket was a packaging trap wearing a large-logo costume.
Recommendation that changes. Do not staff a generic enterprise hunter pod. Staff two application engineers and one technical seller against structural accounts in the 50–250 and 250+ bands, with a kill list for packaging RFPs that clear less than $4,000 contribution after bid cost. Move craft to distributor-only. The 95 large packaging plants everyone wanted to toast contribute less per logo than a mid-size structural supplier.
Reproduce the arithmetic: if you mis-allocate eight sellers to packaging enterprise at $28k cost-to-serve each, you burn roughly $224k of coverage cost chasing $2,940 contribution accounts. Put those eight on structural and the same coverage cost sits against $17,540 contribution accounts — about 6× the contribution per logo before you even count lower churn.
That is what "segments behave differently on the driver" looks like in numbers. The driver here was contribution after cost-to-serve and retention, not headcount.
Ned's rule. If switching two customers between your segments would not change price, service model, or chase priority, erase the line. The interviewer is grading the commercial difference, not the labels.
How to run the cut in forty-five seconds
- Ask what decision you are supporting (price, invest, kill, channel).
- Name the one driver that would change that decision.
- Propose one observable split that should move the driver — prefer need or usage over demographics unless the case gives you a tight proxy.
- Sketch a three-row table: volume, price or margin, cost-to-serve or churn.
- State who you will not serve with the expensive motion.
If the exhibit already has demographics, treat them as a hypothesis to falsify, not as the answer. Re-cut once. The flip is the insight.
Practice this today
Take any profitability or growth case and force two segmentations: one demographic, one needs-based. Score yourself only on whether the recommendation changes. If it does not, your needs cut was decorative.
Run a timed structure rep on CoachNed — everything is open for seven days, no card; then $120 for a recruiting season or $49 a month.
Or open the structure drill directly, or take a full case from /consulting-cases and write the kill list before you synthesize.
CoachNed is not affiliated with McKinsey, BCG, Bain, Deloitte, or any firm named in interview folklore.
Frequently asked questions
Is demographic segmentation ever correct in a case?
Yes, when demographics are a tight proxy for the driver — for example, clinic size when implementation cost and ACV both scale with clinician count. State the proxy explicitly: "I am using headcount as a stand-in for implementation complexity." If the proxy breaks when you check margin or churn, re-cut.
What is the difference between needs-based and behavioral segmentation?
Needs-based groups by the job or benefit sought (why they buy). Behavioral groups by observed actions (what they already did: channel, frequency, tenure). Behavior is useful when the exhibit gives usage data; it becomes circular when you define "loyal customers" by retention and then "explain" retention with loyalty.
How many segments should I use in a case interview?
Usually two or three on one primary axis. A second axis is allowed only if the first still mixes two businesses. Four simultaneous cuts is a spreadsheet, not a structure the interviewer can follow in real time.
Should I mention STP or classic marketing frameworks out loud?
No. Say what you will split on and what you will do differently. Naming STP, 4Ps, or a textbook model rarely adds points and sometimes signals you are pattern-matching instead of solving.
When is segmentation the wrong first tool?
Demographic and firmographic variables are easy to observe. Age. Income.
When profit fell and you have not isolated the driver — start with a profitability bridge. When the constraint is capacity or operations and the buyer set is already fixed. When there is a single buyer (a city concession, a regulator). Segment later if mix still matters after the driver is clear.?
Prompt. Northline sells specialty adhesives into manufacturing. Book of 2,100 accounts. Growth is flat.
Sources
- Product Differentiation and Market Segmentation as Alternative Marketing Strategies (Wendell R. Smith, Journal of Marketing, 1956) — foundational framing of segmentation vs differentiation under heterogeneous demand. Checked 2026-09-24.
- Benefit Segmentation: A Decision-oriented Research Tool (Russell I. Haley, Journal of Marketing, 1968) — benefits sought as causal predictors vs descriptive demographics. Checked 2026-09-24.
- Rediscovering Market Segmentation (Daniel Yankelovich and David Meer, Harvard Business Review, 2006) — nondemographic segmentation's purpose: unmet needs and changeable behavior, not ad casting. Checked 2026-09-24.
- New Criteria for Market Segmentation (Daniel Yankelovich, Harvard Business Review, 1964) — early argument that demographics alone are insufficient for marketing strategy. Checked 2026-09-24.