Cohort Analysis groups users by something they share, most often the date they installed, then follows each group over time instead of lumping everyone into one blended average. A cohort might be every user who installed in the first week of March, or every user who reached level five. Watching that fixed group age tells you what a single average never can.

How it works

You define the cohort by a shared trait, set a starting point, then measure a behavior at intervals from that point: day one, day seven, day thirty, and so on. Because the group is fixed, any change you see comes from the users themselves, not from new installs washing into the numbers. Plot several cohorts side by side and patterns appear, such as a retention curve that flattens after a week or a monetization line that keeps climbing for your best groups. Analysts usually lay this out as a grid, with each cohort on one row and the days since install running across the columns, so a single view shows how every group decays or grows relative to the others.

Acquisition cohorts vs. behavioral cohorts

Two groupings do most of the work. An acquisition cohort is defined by when or how users arrived, which is ideal for comparing channels and install weeks. A behavioral cohort is defined by something users did, such as completing onboarding or making a first purchase. Acquisition cohorts answer where value comes from, and behavioral cohorts answer which actions predict it. Read together, they show both the source and the habits of your most valuable users.

Why publishers rely on cohorts

Blended averages hide more than they reveal because they mix brand-new users with loyal veterans. Cohorts separate them, so you can see true Retention Rate curves, compare LTV across install months, and judge whether a feature change actually helped the users who experienced it. Cohorts are also how you find your Golden Cohort, the group that retains and monetizes far above the rest, and then chase more users who look like it.

Measuring an offerwall with cohorts

Cohorts are the cleanest way to measure what an offerwall adds. Compare a cohort that was shown the offerwall against a similar cohort that was not, and track ARPDAU and retention for both over the same window. The gap between them is the offerwall's incremental lift, isolated from seasonality and traffic mix. This is far more reliable than a single before-and-after number, which blends too many moving parts to trust. It also lets you watch whether that lift holds up as each cohort ages, rather than assuming a day-one bump lasts.

Common mistakes to avoid

• Cohorts that are too small. A handful of users produces noisy curves that swing on individual behavior.

• Comparing cohorts of different ages. A day-three group will always look better than a day-thirty group, so align the windows.

• Reading averages instead of the curve. The shape over time carries the insight, not a single blended figure.