What a retention average is hiding

The average user does not exist

A single curve is useful for orientation, but weak for diagnosis. It blends customers who arrived for different jobs, entered through different promises, and encountered different versions of the product. Before acting on the average, ask which mixture changed.

Choose segments with a reason

Do not create dozens of cuts and hunt for a surprising line. Form a hypothesis about why behavior may differ. Acquisition source can reflect a promise; first-session action can reflect intent; geography can reflect availability or language. Each segment should suggest a product decision if the difference is real.

Check the mechanics

App updates, identity stitching, changing notification rules, and late events can reshape retention without customer behavior changing. Compare instrumentation and release timelines before writing a narrative. Small cohorts also need uncertainty, not confident conclusions.

Move from curve to experience

Once a meaningful segment appears, inspect the journey around the divergence and pair the pattern with qualitative evidence. Retention analysis earns its value when it identifies an experience the team can improve or a promise it should make more honestly.

---

Have a related analytics question? Tell us what your team is working through.