Retention curve
On this page
Definition
A retention curve is the share of users still active as a function of time since start. It shows a trajectory rather than a single number: how fast the bulk churns and whether the curve flattens into a plateau. The shape distinguishes a product with real habit formation from a one-off one.
How to compute
For a cohort, compute for each period and join them into a line, often on a log scale so tail differences are visible. To compare cohorts, normalise curves by age and overlay them. A plateau can be estimated by fitting a decaying function and extrapolating its asymptote.
Pitfalls
Curves cannot be compared at different lengths or without normalisation. A plateau means part of the audience stayed, not that retention is high. The early segment is the noisiest. Extrapolation without a model misleads: real curves do not always decay monotonically.