Segments must change decisions.
If two customer groups receive the same message, offer and cadence, giving them different segment names creates little value. Segmentation should start with a simple question: what will we do differently for this group?
That forces the team to connect data with action rather than creating another report.
Use the real purchase cycle.
At-risk should not be defined by an arbitrary number of days. For a product repurchased every 30 days, 45 days may matter; for a six-month purchase cycle, the same threshold is meaningless.
Cohorts and time-to-repeat distributions provide better lifecycle rules.
Lifecycle can vary by category or market.
A customer may be active in category A and dormant in category B. In multi-country businesses, behavior and channel preference can also vary significantly.
The initial model does not need extreme complexity, but it should be able to evolve with the business.
Track movement, not only snapshots.
The key question is how customers move between states, how quickly they move and whether interventions change that movement. Retention is dynamic—not a one-time labeling exercise.