Most segmentation decks describe customers. The useful ones predict them. A segment earns its keep only when knowing which one a person belongs to tells you something you can act on, what they'll buy, what they'll pay, what would make them switch. The three families below differ entirely in how well they do that, and the most common mistake is reaching for the easiest cut rather than the one that actually moves a decision.

The quick version

  • Demographic segmentation groups people by who they are (age, income, region, company size). Easy to measure, easy to reach, but often a weak predictor of what they'll actually do.
  • Behavioural segmentation groups by what people do (heavy vs light use, channel, loyalty, occasion). It predicts well, but you can only see it after people start behaving.
  • Needs-based segmentation groups by the underlying job or outcome people are trying to achieve. It's the most useful for product and proposition decisions, and the most work to build.
  • Pick the cut by the decision in front of you, not by the data that's easiest to pull. A segment you can't act on differently is just a description.

The idea in depth

The term arrived with a single 1956 paper. Wendell Smith, writing in the Journal of Marketing, drew the line between product differentiation (bending demand toward what you make) and market segmentation, "viewing a heterogeneous market… as a number of smaller homogeneous markets in response to differing preferences." His insight was that markets are lumpy. Rather than average everyone into one bland offer, you can serve distinct clusters precisely (Smith, 1956). Nearly seventy years on, that's still the whole game: find clusters that differ in a way that matters, then treat them differently.

The textbook map, the one in Kotler and Armstrong's Principles of Marketing, sorts the bases into four: geographic, demographic, psychographic and behavioural. For working purposes it's cleaner to collapse those into three lenses by how far they sit from the actual buying decision.

flowchart TD
    A(["A heterogeneous market"]) --> B(["Demographic: who they are"])
    A --> C(["Behavioural: what they do"])
    A --> D(["Needs-based: what they're trying to get done"])
    B --> E(["Easy to measure & reach
weaker predictor"])
    C --> F(["Strong predictor
visible only after the fact"])
    D --> G(["Best for product decisions
hardest to build"])
The three lenses, ordered by how close they sit to the buying decision. Leaders Loop

Demographic: the cut you can always make

Age, income, gender, location; in B2B, the equivalent "firmographics", industry, company size, region. Demographics win on practicality. They're recorded everywhere, ad platforms let you target them directly, and two people in the same bracket are easy to find again. That's not nothing, and for some categories, pensions, children's products, anything regulated by age, the demographic is the need.

The limitation is that demographics describe the buyer's circumstances, not their motivation. Two 45-year-olds on the same salary in the same postcode can want opposite things from the same product. Treating a demographic as a proxy for a preference is where a lot of segmentation quietly fails. So the move is to use demographics for reach and sizing, but to test before you assume they predict behaviour: pull last year's data and check whether the bracket actually separates buyers from non-buyers. If the lines cross, the demographic is a label, not a lever.

Behavioural: what people actually did

Heavy versus light users, new versus loyal, the channel they buy through, the occasion that triggers a purchase. Behavioural data has an obvious virtue: it's not a guess about what someone might do, it's a record of what they did. The classic version is the 80/20 split, a minority of heavy users often drives the bulk of volume, and they may want, and be worth, something different from everyone else.

The catch is timing. You can only segment on behaviour you've already observed, which makes it strong for retaining and growing existing customers and weak for a market you haven't entered yet. It also tells you what happened without telling you why, two customers can show the identical purchase pattern for completely different reasons, and a tactic that delights one will annoy the other. The fix is to layer behaviour onto a needs read rather than treat it as the whole story: start from your usage data to find the high-value patterns, then interview a handful of those customers to learn the motivation underneath. Behaviour tells you where to look; it doesn't tell you what you found.

Needs-based: what they're trying to get done

The most decision-useful cut groups people by the underlying outcome they're hiring a product to deliver. Clayton Christensen's Jobs-to-be-Done framing is the sharpest version: a fast-food chain trying to sell more milkshakes got nowhere tuning flavour and price by demographic, until it noticed that a large share sold before 9am to lone commuters. They were "hiring" the shake for a job, something filling, one-handed and slow to finish on a dull drive. Its real competitors were bananas and bagels, not other milkshakes. The job, not the demographic, was the segment.

A segment you can't act on differently is just a description with a chart on it.

Needs-based segments travel further than demographic ones because the need is stable while the demographics around it vary. The honest limitation is cost and fragility: needs aren't sitting in your CRM, so you build these segments through qualitative research and analysis, and a clumsy version can produce tidy "personas" that flatter the deck and predict nothing. That risk is exactly what Daniel Yankelovich, who introduced non-demographic segmentation back in 1964, and David Meer warned about in Harvard Business Review: plenty of companies run an expensive segmentation exercise, then can't point to a business decision it changed (Yankelovich & Meer, 2006). Their fix is a discipline worth stealing: segment on what people do and need, not just who they are, and only keep a segment if you can act on it differently. Set the bar before you start: every proposed segment has to change at least one real decision, a feature, a price, a channel, a message. If it changes nothing, delete it.

A worked example

Picture a mid-market gym chain deciding where to invest next year. (Figures below are illustrative.) The marketing team's first map is demographic: members split into "under-30s" and "over-40s." It's a clean slide and it explains almost nothing, both groups churn at roughly the same rate and buy roughly the same plans.

So they re-cut on behaviour, using twelve months of entry-scan data. Now a real pattern appears: about 20% of members visit four-plus times a week and renew reliably; a larger group visits once a fortnight and churns inside six months. That's actionable, but it still doesn't say why the fortnightly group is drifting, only that they are.

A dozen interviews with churned members surface the need. The frequent visitors are there for identity and progress; the drifters joined for a specific, time-bound outcome, "get fit before the wedding," "back pain my physio flagged", and once the deadline passed or the goal slipped, the membership had no job left to do. Three lenses, three different pictures. The demographic cut would have justified a generic ad campaign. The needs cut points somewhere sharper: a structured, outcome-dated programme for the "deadline" segment, with check-ins that keep the job alive past the original date.

flowchart LR
    A(["Same gym members"]) --> B(["Demographic cut
under-30 / over-40
→ no real difference"])
    A --> C(["Behavioural cut
frequent / drifting
→ who is at risk"])
    A --> D(["Needs cut
identity / deadline
→ why, and what to build"])
The same members, three lenses, each later cut answers a question the previous one couldn't. Illustrative. Leaders Loop

The point isn't that needs always beat demographics. It's that you read the demographics for sizing and reach, the behaviour for who to prioritise, and the need for what to actually build, and you don't confuse one for another. This is the same posture customer needs identification formalises: the most valuable segment is often defined by a need the customer can't yet articulate.

Frequently asked questions

Which type of segmentation should I start with?

Start from the decision, not the data. If you're sizing a market or buying media, demographics and firmographics are the right tool. If you're deciding who to retain or upsell, behaviour wins. If you're shaping a product or proposition, you need the underlying need. The failure mode is picking the cut that's easiest to pull and then forcing your decision to fit it.

Aren't personas the same as segments?

No, and conflating them is where segmentation goes soft. A segment is a group that differs in a way that changes what you do; a persona is a narrative character you build to make a segment memorable to a team. Personas are a communication device on top of a segment. If you write the persona first and the segment never existed, you've drawn a cartoon, not a strategy.

How many segments should I have?

As few as the decision allows. Every extra segment costs you a distinct tactic to serve it, so the test is operational: can you genuinely treat each one differently? Most organisations can act on three to five at a time. A 12-segment model that the business serves identically is a research artefact, not a plan.

Does segmentation even work? I've seen it dismissed.

The scepticism is earned, not the principle. The research records plenty of expensive exercises that produced little of value, usually because the segments were descriptive (who people are) rather than actionable (what they need and do). Segmentation built on a real difference that changes a real decision still pays; segmentation built to fill a slide doesn't. The discipline is in the bar you set, not the technique.

How is B2B segmentation different?

The lenses are the same, but the demographic layer becomes firmographic (industry, size, geography) and you have to account for a buying group rather than one buyer. Bonoma and Shapiro's "nested" approach is the standard frame: work from the easy outer layers (firmographics) inward to the hardest, most valuable ones (the individual buyers' situations and personal characteristics), going only as deep as the decision needs.

Related in the Toolkit

Where to go next