Retention
The Retention page shows how many of your new users keep using the extension. Open it from the sidebar under any extension. It works the same for the SDK and the WXT module, using only the automatic events — you do not need to track anything extra.
What the page shows
- Day 1, day 7 and day 30 retention — the share of new users who were active on exactly that day after they installed. Each number shows how many users it is based on.
- The cohort table — one row per install week, with its number of new users. Column Week N is the share of that week's users who were active on any day of the Nth week after their install week. Week 0 is the install week itself, so it is always 100%.
- Range —
4w,8w,12w(the default) or26w: how many install weeks to show, ending with the current one. The table goes up to week 12. Day 30 needs8wor longer: no install day in a 4-week range is 31 days old yet, so on4wday 30 always reads—.
Definitions
All dates are UTC, and weeks start on Monday.
| Term | Meaning |
|---|---|
| Install day | The date of a user's first __first_open or __install event, whichever came first. Users with neither event are left out. A user whose first install was before the range is not counted as new, even if they reinstall during it. |
| Active | At least one event that day — any event, __daily_ping included. __uninstall is not activity. |
| Day N retention | Users active on install day + N, divided by the users who installed. |
| Week N retention | Users active on any day of install week + N, divided by the users who installed that week. |
Readings that are not in yet
Numbers only count time that has fully passed:
- Day N only includes install days whose day N has ended. Users who installed recently are left out rather than counted as not retained, and the number reads
—until some install day in the range qualifies — day 30 needs installs from at least 31 days ago. On the4wrange that never happens, so the tile asks for a longer range instead. - The current week of each row is hatched and left blank (Week not over yet in the legend) instead of showing a partial percentage. It fills in after Sunday.
- The newest row reads This week while its install week is still under way, because new users are still joining it.
When the numbers look off
- Over your plan limit. Once an organization passes its monthly event limit, only
__install,__updateand__uninstallare accepted, so__daily_pingand your own events are dropped and retention reads low for those days. The page marks this with an Incomplete chip. - Adding Moderok to an extension that already has users. Every existing user gets a new anonymous id and fires
__first_openthe first time the updated extension runs, so they appear as new installs in that week. - School Chromebooks. Managed ChromeOS devices in ephemeral mode wipe the profile at logout, so the same student shows up as a new user who never comes back, again and again. That pulls retention down. The Trends Installs metric filters these out; Retention does not, because a filter that drops users who never returned would make every number look better than it is. See Installs vs Chrome Web Store for the details.
- Cohorts vs the Installs metric. The two can differ in either direction. Installs counts every
__installevent, reinstalls and repeat installs included, and leaves out the ChromeOS sessions above. A cohort counts each user once, in the week of their first-ever__first_openor__install, and keeps those ChromeOS users.
