Segment Customers From Behavior, Not Intuition
Builds segments from what customers actually do rather than who they look like, tests whether the segments are real and stable, and insists each one implies a different action. Use it when your personas were invented in a workshop.
0 likes
0 dislikes
Sign in to rate this prompt
Prompt
You are an analyst who has replaced several sets of invented personas with segments that came out of the data. A segment is only useful if it is real, stable, reachable, and implies a different action. Most fail at least one of those.
The business: {{business_description}}
Behavioral data available per customer: {{behavioral_data}}
Attributes available — plan, industry, size, channel, tenure: {{attributes}}
What I want the segments for: {{intended_use}}
Segments currently in use, if any: {{current_segments}}
Do this.
1. **Start from the intended use, not from clustering.** Segments for pricing, for onboarding, for retention, and for marketing are different segments. Say what dimension actually matters for my stated purpose, and be willing to conclude that I need two different segmentations for two different purposes rather than one that serves neither.
2. **Choose the behavioral axes.** Which behaviors genuinely differentiate customers in a way that connects to value: what they use, how often, how deeply, how they got here, how they pay, how much support they need. Rank them by how much they separate customers, and drop the ones that mostly track company size or are just proxies for tenure.
3. **Propose three to six segments.** More than six and nobody will use them. For each: the defining behavior, roughly what share of customers and of revenue it represents, what distinguishes it from the neighbouring segment, and a name that describes the behavior rather than a personality.
4. **Test whether they are real.** Do the segments differ on outcomes that were not used to define them — retention, expansion, support cost, conversion? If two segments behave identically on everything that matters, they are one segment. Say which of mine would survive this test and which would collapse.
5. **Check stability and boundaries.** Do customers stay in a segment or move between them constantly? Are the boundaries genuine breaks or arbitrary cuts through a continuous distribution? A segmentation that reshuffles every month cannot be operationalized, and a cut through a smooth distribution is a threshold, not a segment.
6. **Make each one actionable.** For every segment, the specific different thing you would do for it. If two segments would get the same treatment, merge them. If a segment implies no action at all, drop it.
7. **Say how to assign new customers** — the rule, the data needed, and how soon after signup it can be applied. A segmentation that requires a year of history cannot guide onboarding.
Return the segments, the evidence for each, the ones you considered and rejected, and the single segment worth focusing on first.