What is segmentation?

The main types

• Behavioural. Grouped by what people did — the most predictive in most products.
• Demographic. Age, location, language. Easy to obtain and usually weakly predictive on its own.
• Value-based. By spend or predicted LTV, which is what most commercial decisions actually need.
• Lifecycle. New, active, at-risk, lapsed — the segmentation that drives most lifecycle messaging.
• Needs-based. Grouped by what the user is trying to accomplish, which is the hardest to build and the most useful.

What makes a segment worth having

A segment is useful only if it changes what you do. Three tests are worth applying before building one.

Is it distinct? Members should behave measurably differently from non-members, or the division is arbitrary.

Is it actionable? You must be able to reach the segment and do something different for it.

Is it stable enough? A segment whose membership churns weekly cannot support a strategy — though it can support a real-time decision, which is a different use.

Most segmentation projects produce elegant taxonomies that fail the second test, and are quietly abandoned.

Segmentation on conversational surfaces

The usual model — build a persistent segment, store it, target it later — depends on recognising the same person over time, which conversational surfaces do not offer.

What replaces it is situational rather than persistent: the segment is defined by what the user is doing in this conversation, evaluated at request time and gone afterwards. See synthetic audience.

The practical translation from a segment brief is usually straightforward once the habit shifts: describe the situation rather than the person. 'Small business owners aged 30–50' becomes 'someone choosing a business bank account for a company with under ten employees' — more precise, and directly matchable.