You have a survey with a few thousand answers and a leadership team that wants to know which customers to focus on. Segmentation is the step that turns one large, mixed sample into groups you can describe, size and target. It sits near the end of the market research process, and the type you choose shapes which questions you must ask at the start.
Key takeaways
- The four main types of segmentation in market research are demographic, geographic, psychographic and behavioral, defined by who people are, where they are, what they value and what they do.
- Demographic and geographic segments are easiest to build and reach, while psychographic and behavioral segments explain motivation and action more directly but need more survey design.
- Segments are built either a priori, by choosing the grouping variable first, or data-driven, by clustering respondents on attitude or need ratings.
- A good segmentation passes three tests: each segment is big enough to matter, differs on the answers that count, and can be found and acted on.
What are the four main types of segmentation in market research?
The four types differ in what they use to define a group. Each maps to a different kind of survey question, as the table shows.
| Type | Groups people by | Example survey question | Example segments |
|---|---|---|---|
| Demographic | Who they are | ”What is your age?” “What is your household income?” | Adults 18 to 34; households over a set income |
| Geographic | Where they are | ”What is your ZIP code?” “Do you live in a city, suburb or rural area?” | Northeast urban; rural Midwest |
| Psychographic | What they believe and value | ”How much do you agree: I like to try new products before my friends do?” (1 to 5 scale) | Early adopters; value seekers; brand loyalists |
| Behavioral | What they do | ”How many times did you buy this product in the last 3 months?” | Heavy, light and lapsed buyers |
Demographic and geographic segments are the easiest to build because the answers are factual and often already in your customer file. They are also the easiest to reach, since media and sales territories are organized around them. The weakness is that two people of the same age and ZIP code can want very different things.
Psychographic and behavioral segments explain motivation and action more directly, but they take more survey design to measure. Behavioral questions need a clear time frame, and psychographic items usually need several statements on a rating scale.
Older research texts name the four types differently: a priori, usage, attitudinal and needs-based. They overlap with the list above. A priori covers demographic and geographic groups, usage is behavioral, attitudinal is psychographic, and needs-based groups people by the benefits they are looking for.
How are segments built from survey data?
There are two routes, and they answer different questions.
What is a priori segmentation?
In an a priori segmentation, you decide the groups before looking at the data, using a variable you already believe matters. You might split respondents into three age bands or into customers and non-customers, then compare their answers. This is deductive reasoning, and it is the most common approach because it is quick, cheap and easy to explain. The risk is in the assumption: a rule that held ten years ago, such as older adults being slower to adopt new devices, may not hold now.
Check it against your data. Our guide to deduction and induction in market research covers when each way of reasoning fits.
Usage segmentation is a special case that works well alongside it. You rank customers by how much they buy and cut the list, either into ten equal deciles or into the top 20% and the rest (the Pareto approach).
Here is an illustrative example. A survey of 1,000 customers records 5,000 purchases in a year. The heaviest 200 buyers (the top 20%) account for 3,400 of them, which is 68%, or 17 purchases each. The other 800 buyers account for 1,600 purchases, or 2 each. That gap tells you where retention effort pays off, and you can then profile the heavy group by age, region or attitude.
What is data-driven segmentation?
In a data-driven segmentation, you let an algorithm find the groups. You choose the survey questions that matter, usually attitude, preference or need ratings, and a clustering method sorts respondents so that people in the same cluster answer alike and people in different clusters answer differently. Common methods are k-means and hierarchical clustering; our explainer on hierarchical clustering walks through one with a small example.
The inputs decide the result. Demographic variables make poor clustering inputs because they rarely separate people by what they want. Rating scales, such as importance of price, quality or convenience, work better.
Needs-based segmentation applies the same idea: measure what people value, using importance ratings, MaxDiff or conjoint analysis, then cluster on those scores to find groups with different priorities. The point is to build products around those needs rather than to sell products you have already made.
How do you know if a segmentation is any good?
Statistics will give you segments whether or not they are useful, so test each solution against three questions.
- Size. Is each segment big enough to matter commercially? A segment of 3% of the market may be fascinating and still not worth a dedicated campaign. Also check the sample: very small segments produce unstable percentages.
- Distinctness. Do the segments differ on the answers that matter, and do those differences hold up in significance tests? If two segments have nearly the same profile, merge them.
- Actionability. Can you find these people and do something different for them? A segment you cannot identify in your customer file or reach through a channel is a description, not a plan.
Add a fourth check when you can: stability. Re-run the analysis on a random half of the sample, and see whether the same groups appear.
What mistakes do people make with segmentation?
- Starting with demographics because they are easy. Age and income describe who people are, not what they want. Use them to profile segments, not always to define them.
- Clustering on too many variables. Feeding in 60 questions adds noise. Select the handful that relate to your business decision first.
- Choosing the number of segments for convenience. Six segments is not better than four because it feels thorough. Pick the number your team can brief and measure.
- Treating segments as people. A segment is a statistical group, and individual members will not fit the profile exactly. Name segments for what they do or want, not for a stereotype.
- Ignoring the business goal. The right type depends on the decision and the budget. Cost and the goals of the decision makers often decide which type you use, so settle the goal before the method.
How do you do this in practice?
Once segments exist, you need to profile them: a crosstab of each segment against demographics, usage and key attitudes, with weighting and significance testing, so you can see which differences are real. Halo Reports produces crosstabs and banner tables with weighting and significance testing, plus charts and interactive report pages, so a segment profile can be shared with the people who will act on it. Segments also earn their keep when they carry forward: connecting findings across studies and waves lets teams reuse what they know about each group instead of starting over, and you can track brand health and competitors continuously after the segmentation study ends.
Start with the decision. Write down what you want the segments to change, such as a product line, a message or a channel, then work back to the questions that will measure it. If you are earlier in the process, read the first step in the marketing research process before you design the survey.