Build one segmentation in front of the mirror. A detergent maker starting from demographics gets 'housewives 25–45, SEC B', a media plan, not an insight. Starting from needs gets segments like 'stain-fighting for school uniforms, price-per-wash guardians, fragrance-led freshness seekers, machine-protection worriers', each implying a different product, claim, and price. The craft sequence: qualitative work surfaces the needs and occasions, quantitative work clusters buyers by those needs and sizes each cluster, then demographics and media behaviour are laid over each cluster so it can be found and bought against. Needs make segments meaningful. Demographics make them reachable. You need both layers, in that order.
Then every candidate segment faces the five Beginner checks plus the two scoring axes. Attractiveness: volume and value today, growth trajectory, intensity of competition already serving it, price realisation it supports. Ability to win: do your brand's existing associations, distribution, cost structure, and product truths give you a credible right to play, or would winning require becoming a different company? Score them 1–5 with named evidence per cell and the map draws itself: attack where high meets high, make selective bets where attractiveness is high but your hand is weak (the cell that demands investment before revenue), harvest the winnable-but-small, and write down, explicitly, who you will not serve. The unserved list is what makes it strategy.
Targeting patterns follow from the grid. Concentrated targeting (one segment, everything committed) is the challenger's default: depth beats breadth when resources are thin, and owning one segment's loyalty funds the next. Differentiated targeting (several segments, distinct mixes) is the scale player's game, and its tax is real: separate products, campaigns, sometimes separate channels. Run the HUL detergent ladder in your head (value, mid, premium as distinct brands) and price what that costs before admiring it. Mass marketing survives where needs genuinely don't differ, and as the Ehrenberg-Bass camp insists, in many repertoire categories reach across the whole category buyer base out-earns precision aiming. The contested ground is honest: segmentation is at its most defensible for offer design (what to build, what to promise, what to charge), and at its most debatable as a media-narrowing device.
The pathologies are as instructive as the method. Cluster mysticism: statistically elegant segments ('the Balanced Optimisers') that no media plan can find and no salesperson recognises, failing both accessible and actionable. Persona theatre: fictional biographies with stock photos standing in for evidence. Over-segmentation: twelve micro-segments for a company with one production line and two campaigns a year. Static maps: segments built in 2020 still steering 2026 budgets while the category re-sorted itself. The tell of a working segmentation is that operations can name which segment an actual customer belongs to, from their behaviour, within a minute.