Retention
See whether visitors come back
Acquisition numbers rise for a while whatever you do. Retention is the number that says whether any of it stuck.
Five weekly cohorts, 1,740 people in the newest. The shape is triangular because the newest has had the least time to return.
01
Across a row is the life of one cohort
A cohort is everyone who arrived in the same week. Week 0 is the week they arrived, so it is always 100%, and the rate after that is the share of that group seen again later. The first row falls from 100% to 42% in a week, then flattens near 25%. That flattening matters more than the first drop: a curve that settles has a real core audience, while one that keeps falling toward zero does not.
- Week 142%
- Week 231%
- Week 327%
- Week 426%
- Week 525%
02
Down a column is the only fair comparison between cohorts
Comparing groups of different ages on their totals says nothing; comparing them at the same week of life says everything. Week 1 moves from 42% in the oldest cohort to 51% in the newest, and later cohorts holding up better than earlier ones is the signal that something you shipped in between worked. The averages below the table only count cohorts that have actually reached each week, so a young cohort never drags the number down by being young.
- Week of 4 Aug42%
- Week of 11 Aug44%
- Week of 18 Aug46%
- Week of 25 Aug49%
- Week of 1 Sep51%
- Average46%
03
Retention is mostly an acquisition question
The same product retains very differently depending on where people came from. Newsletter readers hold at 48% by week three while social referrals fall to 8%. Averaged together those become a single unremarkable curve that suggests fixing the product; split apart, they suggest changing where the traffic comes from. Splitting the curve by source usually explains more than any change to the product would.
- Newsletter48%
- Organic search29%
- Social referral8%
What a cohort table cannot show yet
The newest cohort has only had one week in which to return, so the table is triangular by construction and the right-hand columns are always the thinnest evidence on the page. Retention is also measured for the group rather than by following a named individual: visitors are counted without a persistent identifier, so "came back" is a property of the cohort, not a person you could look up.
Questions about retention
These are on the guides index too, alongside every other question the site answers.
What is a cohort?
The group of people who arrived in the same period, usually a week. Following each group separately shows whether the product is getting stickier, which an overall active count hides completely.
What counts as a returning visitor?
Someone from the cohort who came back in a later week. Because visitors are counted without a persistent identifier, this is measured at the level of the group rather than by following a named individual.
Why does the newest cohort have so few columns?
A cohort that arrived last week has only had one week in which to return. The triangular shape is expected, and it is why the averages below the table only include cohorts that have actually reached each week.
Related
See your own cohorts
Weekly cohorts build themselves from the pageviews you are already sending. There is nothing to define first.
14-day free trial. No credit card required.