Retention
Cohort tables that show whether people come back. Group visitors by when they first arrived and see what share returns each week after.
Reading the table
The retention view groups visitors into cohorts by the week they first showed up, then tracks what percentage returned in each following week. Rows are cohorts, columns are weeks since first visit. A healthy product shows the curve flattening after the early drop, a leaky one trends toward zero.
Anonymous versus identified
Retention works for anonymous visitors out of the box, using the device id. It gets meaningfully better once you identify users, because the same person on two devices stops counting as two visitors who each churned.
What to do with it
Retention is the honest metric behind most vanity ones. A launch that spikes traffic but leaves week 1 retention untouched brought the wrong audience. Compare cohorts from before and after a change to see whether it moved actual stickiness, and ask the AI assistant to summarize the movement if you want it in words rather than a matrix.