Retention glossary
Plain definitions for the terms used on this site.
The words you need to read a cohort retention table: cohorts, Day 0 and Day N, the three retention types, bounded and unbounded numbers, diagonals, mix shifts and more, each in one or two plain sentences.
What do the retention terms mean?
Install cohort: A group of users who installed on the same day (or week). The unit every retention number is about.
Day 0: The install day. Activity on Day 0 never counts as a return.
Day N: The Nth day after install, counted in calendar days in one time zone or in 24-hour windows; say which.
Return event: What counts as being active: usually any app open. Some teams use a meaningful action instead.
Classic retention: The share of a cohort active on exactly Day N. Also called N-day, Day-N or exact-day retention. Bounded.
N-day retention: Another name for classic retention.
Rolling retention: The share of a cohort active on Day N or any later observed day. Also called unbounded retention. Keeps rising as time passes.
Unbounded retention: Another name for rolling retention, after its open-ended window.
Cumulative retention: The share of a cohort active at least once between Day 1 and Day N. Bounded, never falls as N grows.
Bracket retention: The share of a cohort active at least once inside a range of days, such as Days 8 to 14. Cumulative is a bracket that always starts at Day 1.
Bounded: A retention number that is fixed once its window has passed. Classic and cumulative are bounded; rolling is not.
Mature cell: A cell whose Day N has fully passed for that cohort. Only mature cells should be compared.
As-of date: The last day included in the data. Rolling numbers only mean something next to it.
Cohort table: Install cohorts as rows, days since install as columns, retention in the cells. Usually a triangle, because newer cohorts have fewer days.
Diagonal: Cells in a cohort table that share one calendar date. A bright or dark diagonal points to an event on that day.
Retention curve: One cohort's (or a pooled set's) retention plotted against Day N.
Flattening point: Where a classic retention curve stops falling and levels off at its habitual users.
Blended retention: Retention across all channels together. Moves when the channel mix moves, even if no channel changed.
Mix shift (Simpson's paradox): A blended number moving the opposite way to, or without, its parts because the share of each part changed.
Churn rate: The share of users who stop using the app over a period. Roughly the flip side of retention, but defined in many ways; check which.
Stickiness (DAU/MAU): Daily active users divided by monthly active users: how often active users come back. A complement to cohort retention, not a replacement.