Customer Segmentation 2026: Does the Answer Change?

The test a customer segmentation has to pass

Most segmentation advice says to choose an axis that is meaningful, measurable and reachable. Those are necessary and none of them is sufficient, because an axis can be all three and still leave the recommendation unchanged.

The operational test is one sentence. If the two groups would receive the same recommendation, they are one group for the purposes of this case.

That test has a useful second edge. Running the numbers for two groups and finding the same answer is a finding worth saying out loud, because it tells the room the axis is wrong and that you checked. Oliver Wyman lists evaluating and using data, and asking the right questions, among its six published evaluation criteria, checked 30 September 2026 at oliverwyman.com.

Age, region and company size are descriptions until you have shown they predict something. Predicting something, in a case, means moving a number that is already on the table.

Two ticket classes in a published case interview

Oliver Wyman publishes a case in which a prospective entrant estimates the daily revenue of an established Dubai water park, checked 30 September 2026 at oliverwyman.com. The chain is fully published and it recomputes.

Four slides, running seven hours from 10 in the morning to 5 in the afternoon, at 100 riders an hour. Seven times 100 is 700 riders per slide per day, and four slides give 2,800 rides. At an assumed two rides a visitor that is 2,800 over 2, so 1,400 visitors.

Tickets are 300 dirhams for visitors over twelve and 250 for those under, with an assumed 70 and 30 split. Ancillary spend is 50 dirhams of food and drink bought by 80 percent of visitors and 20 dirhams of merchandise bought by half.

By hand. 70 percent of 1,400 is 980 visitors and 30 percent is 420. Ticket revenue is 980 times 300, which is 294,000 dirhams, plus 420 times 250, which is 105,000, giving 399,000. Ancillary per visitor is 50 times 0.8 plus 20 times 0.5, so 40 plus 10, which is 50 dirhams. Total ancillary is 1,400 times 50, which is 70,000 dirhams, matching the firm's published 56,000 plus 14,000. The day totals 399,000 plus 70,000, which is 469,000 dirhams, and 469,000 over 1,400 is exactly 335 dirhams a head.

Per segment, the over twelve group yields 300 plus 50, so 350 dirhams, and the under twelve group yields 250 plus 50, so 300. The spread is 50 over 300, which is 16.7 percent. Cross check: 980 times 350 is 343,000 and 420 times 300 is 126,000, and those add back to 469,000.

The variable that moves the answer more than customer segmentation

Now the point of the exercise. The age segmentation is clean, measurable, and it does not change what the entrant should do, because both groups walk into the same park on the same day and are served by the same four slides.

Two assumptions that are not segments at all move the total further.

By hand. At 2.5 rides a visitor, 2,800 rides support 2,800 over 2.5, which is 1,120 visitors, and 1,120 times 335 is 375,200 dirhams. At 70 percent utilization, rides fall to 70 percent of 2,800, which is 1,960, so visitors are 1,960 over 2, which is 980, and 980 times 335 is 328,300 dirhams.

So a 16.7 percent spread between the two customer segments sits next to a 20 and a 30 percent swing from two operating assumptions that no customer axis captures. The case is candid about the second: it states that each slide "operates continuously without downtime" and that capacity limits from other attractions are not included. The published estimate is a capacity ceiling on a peak Saturday in April, and a half full park returns the same 2,800 of capacity and far less money.

That is the shape of a segmentation that describes without deciding. The right handling is to present it in one line, note that both segments receive the same recommendation, and spend the time on utilization instead. The ancillary basket is worth checking for the same reason: 70,000 over 469,000 is 14.9 percent of the day, so it matters less than either operating assumption.

Where customer segmentation did decide a case interview

The contrast is in another published case from the same firm, and it is worth holding next to the first.

In its Indian digital therapy pricing case, the income distribution is published as 50 percent low income, 40 percent middle and 10 percent high, against 10,000,000 autistic children, checked 30 September 2026 at oliverwyman.com. The addressable base after filters is given as 4,200,000, and the answer captures "25% of the calculated user base (25% of 4.2 million, which is roughly 1million)".

That segmentation decides the case, because the surviving count is the denominator of a cost based price. Ten billion rupees of annual fixed cost over about one million users is 10,000 rupees a user, and the published price of 15,000 rupees is that figure with a 50 percent markup.

Change the income cut and the price changes with it. Serving only the high income tenth would leave 1,000,000 children before any other filter, and a 25 percent capture of that is 250,000 users, which puts fixed cost per user at 10,000,000,000 over 250,000, so 40,000 rupees. At a 50 percent markup the price would be 60,000 rupees, which is above the cheapest competitor at 50,000. The segmentation has moved the price by a factor of four and reversed the positioning.

Two published cases, two customer axes, and only one of them is load bearing. What separates them is whether the axis appears in the formula that produces the answer. Axis quality has nothing to do with it: both are clean cuts of a real population.

Deloitte's retail case offers a third variant worth noting: it segments by channel rather than by customer, targeting "an aggressive +40% annual growth in direct-to-consumer digital channels over the next five years", checked 30 September 2026 at deloitte.com. No customer segment appears in it at all.

A four step test you can run in the room

1. Write the formula that produces the answer. Revenue per visitor times visitors, or fixed cost over users, or contribution per account times accounts.
2. Ask whether your proposed axis appears in it. If it changes a term, keep it. If it only relabels the same term, say so and move on.
3. Compute the answer for each group. Two numbers, and if they are within a few percent the groups are one group for this decision.
4. Name the unsegmented variable with the largest swing. Utilization, rides per visitor, renewal rate, filter pass rate. That is usually where the case actually lives.

Step four is the one that pays. In the water park the answer turns on how full the park is, and no customer axis reaches it. In the pricing case the answer turns on the income cut, and the customer axis is the whole model.

Three related pages. The sequence that makes a segmentation into a target and a position is in the STP framework. The mechanics of cutting a population are in segmentation analysis. The version where segments genuinely differ in cost to serve is in customer profitability cases.

What a decorative axis costs

The weights below are ours, taken from src/lib/grading. Firms publish criteria per question in one case and method steps in several; a weighted score is not public anywhere in our files.