Segmentation

Picture this: a chain of fitness clubs runs a satisfaction survey across all its members. The average score is 4.1 out of 5. Pretty decent.

But the moment you break the responses down by segment, the picture changes dramatically. Morning-pass members give 4.6 — they love everything. Evening members give 3.4: crowded gyms, lines for the machines, not enough parking. Members with a personal trainer give 4.8: individual attention, no problems. The 4.1 average masked two opposite realities. Without segmentation, management would never have learned that evening members are on the verge of leaving — and they are the ones generating most of the revenue. Segmentation is the tool that turns an "average across the whole hospital" into a diagnosis for each ward.

What segmentation is

Segmentation is the division of an audience into homogeneous subgroups (segments) based on specific criteria: demographic, behavioral, psychographic, or situational. The goal is to uncover differences between groups that hide behind overall averages and to make targeted decisions for each segment instead of one "one-size-fits-all" decision for everyone.

In marketing, segmentation answers the question "Who should we offer what to?" In surveys, it answers "Who exactly is satisfied, who isn't, and why?" It is not an optional add-on to analysis but a mandatory part of it: data without segmentation is like a map without a scale. You can tell something is somewhere, but you can't tell how far away.

Why you should segment survey data

Spotting hidden problems. Overall satisfaction may be high, yet one segment is critically unhappy. Without segmentation you won't see it. And that very segment may be the most valuable in terms of revenue or strategy.

Targeted actions instead of carpet bombing. "Customers are unhappy with delivery" — so what do you do? Speed it up? Make it cheaper? Expand the coverage area? If you segment: city-center customers are happy (next-day delivery), the regions are unhappy (5–7 days). So the problem isn't delivery in general but regional logistics. The solution is specific and addressed to the right place.

Personalizing communication. Different segments respond to different messages. A younger audience values speed, an older one values reliability. New customers want clear onboarding, long-time customers want advanced features. Segmentation lets you speak to each group in its own language.

Tracking trends by group. Overall NPS rose from 30 to 35. Good. But dig in: NPS for corporate clients climbed from 20 to 45 (you launched a new integration), while NPS for individual clients dropped from 40 to 25 (you raised prices). One trend, two realities.

Types of segmentation

Demographic

Division by objective characteristics: gender, age, income, education, marital status, occupation. The simplest and most common type — because this data is easy to collect and verify.

When it's useful: when analyzing mass consumer surveys, where you need to understand whether the perception of a product differs between men and women, younger and older people, the well-off and the budget-conscious.

Limitations: demographics explain "who" but not "why." Two men of the same age and income may have opposite needs. Demographics are a starting point, not the final answer.

Behavioral

Division by actions: purchase frequency, average order value, features used, recency of the last contact, acquisition channel, lifecycle stage (new / active / dormant / churned).

When it's useful: almost always. Behavior is the best predictor of future actions. A customer who buys once a week and a customer who bought once six months ago are two different types of people with different needs, even if they are demographically identical.

A real-world example. An e-commerce platform segments responses by the RFM model (Recency, Frequency, Monetary). It turns out that "champions" (frequent, expensive purchases) complain about the loyalty program — they want more tangible rewards. The "dormant" group (haven't purchased in a long time) won't come back without a personal discount. The same data, two completely different action plans.

Psychographic

Division by values, interests, lifestyle, and motivation. Harder to collect — it requires special questions — but it gives deeper understanding.

When it's useful: when developing positioning, testing advertising concepts, and building a consumer persona. "Frugal pragmatists" and "impulsive hedonists" are not made-up categories but real psychotypes that determine product choice.

Geographic

Division by place of residence: country, region, city, type of settlement (metropolis / mid-size town / village). Critical for companies with broad geography: the experience of a customer in New York and one in a small town can differ radically — different logistics, different competitors, different levels of service.

Situational (contextual)

Division by the circumstances in which a person interacted with the product. Did they buy online or in a store? Did they contact support by phone or in chat? Do they use it on mobile or desktop? Context shapes the experience, and a survey that ignores context mixes apples with oranges.

How to implement segmentation in surveys

Method 1: Classifier questions in the survey

Add questions at the start or end of the survey that assign the respondent to a segment: "How long have you been using our service?", "Which plan are you on?", "What city are you from?" You'll later use these answers to break the data into subgroups during analysis.

Method 2: Hidden variables

If you already know which segment a respondent belongs to (from your CRM, from a database), pass that information through URL parameters. SurveyNinja has hidden variables for this: data about the segment is automatically attached to every response, and the respondent doesn't have to answer extra questions.

Method 3: Logic jumps

Different segments get different routes within a single survey. For example: new customers get onboarding questions, long-time customers get advanced-feature questions. Logic jumps in SurveyNinja let you build complex routes without creating separate surveys for each segment.

Method 4: Filtering during analysis

Even if the survey is the same for everyone, you can segment the data at the analysis stage. SurveyNinja has built-in response filtering: show only responses from men, only customers on the "Business" plan, only residents of a given city. This lets you compare subgroups without complicating the survey itself.

How many segments to create

It's a common question — and the answer depends on the task and the volume of data.

Too few (2–3). Crude. "Young and old," "men and women" — useful as a starting point, but it doesn't reveal the nuances.

Optimal (4–7). Enough detail while staying manageable. Each segment is large enough for statistically meaningful conclusions and different enough from the others to justify the split.

Too many (10+). Analysis paralysis. With 15 responses per segment, it's statistically meaningless. And it's impossible to develop a separate strategy for 12 subgroups. Rule of thumb: if a segment has fewer than 30–50 responses, merge it with a neighboring one.

Common mistakes

Analyzing only overall averages. "Our CSAT is 4.2" — and so what? Behind that figure there could be a universal "okay" or a gap between the delighted and the disappointed. Always look at the distribution and at the segments.

Segmenting by an irrelevant attribute. Splitting the data by zodiac sign is technically possible. Practically useless. The segmentation criterion must be tied to the business task: if you're looking for the causes of churn, segment by lifecycle stage, not by gender.

Not planning segmentation before launch. If the survey has no classifying questions and no hidden variables are passed, there will be nothing to segment by. Think about segmentation before creating the survey, not after collecting the data.

Drawing conclusions from tiny subgroups. "In the 'men 55+ from a single city' segment there are 8 people, and 6 of them gave a 5" — that's not "an excellent segment score," it's statistical noise. The minimum for meaningful analysis is 30 responses per segment; for serious conclusions, 100+.

Segmentation isn't a complication of analysis — it's the essence of it. Without it, survey data is one number that hides dozens of stories. With it, you get a map that shows who's satisfied, who isn't, who's leaving, and why. The difference between "average satisfaction is 4.1" and "morning members are happy, evening members are leaving" is the difference between a report that gathers dust and a decision that saves revenue.