Analysis
Cohort
Also called user cohort.
Cohort is a group of people who share a starting event in the same period, such as first visit in the week of May 6. You follow the group instead of the whole site.
How it is measured
A cohort has two parts: the entry event and the time bucket. Everyone in the group enters through the same event within the same day, week, or month. The group is fixed once the bucket closes.
Group size and entry rule matter more than the label. Record how many people are in the cohort and how the entry was detected. Without a long-lived identifier, cookieless tools build cohorts from what is visible inside a window, not from a lifetime id.
Worked example
A trail-running store promotes a new shoe in a newsletter on June 3. Of the 1,280 people who first arrived that week, you track those who came back in the next 14 days.
A cohort from a May search campaign has 2,900 people. Comparing the two groups shows the June group returned at 11 percent and the May group at 6 percent, a difference that a sitewide return rate would have blurred.
How it differs
A cohort is the group. Cohort analysis is the act of watching that group over time. The group tells you who is in; the analysis tells you what they did next.
Common errors
Making cohorts too small to read. Changing the entry rule between groups. Mixing weekly and monthly buckets. Treating a segment as a cohort when it has no shared start. Forgetting that late joiners belong to the next bucket.
In practice
Choose one entry event that matters, such as first order or first article read, and one bucket size. Keep both fixed for a quarter. Compare the newest cohort with the two before it.