Definition
Cohort Analysis
Cohort analysis groups users by a shared characteristic - most often the week or month they first used a product - and tracks each group's behavior over time. Instead of one blended average, you see whether users acquired in March retain better than users acquired in January, which isolates the effect of product changes from the effect of who happened to sign up.
Interactive diagram - hover the elements. Plotted values are illustrative examples, not measurements.
How do you run a cohort analysis?
Pick the cohort-defining event (signup, first purchase), pick the return behavior that counts as 'retained' (any session, or a specific value action), choose the time grain (daily for consumer apps, weekly or monthly for B2B), and read the resulting triangle chart: each row is a cohort, each column a period since start. The diagonal tells you about calendar effects; the columns tell you about product effects.
What does a good cohort curve look like?
It flattens. Every product loses users early; healthy products stop losing them - the curve levels into a plateau of habituated users. A curve that decays to zero means no durable habit exists yet, and acquisition spend is refilling a leaking bucket. The plateau height, not the day-1 number, is the figure worth reporting.
Example: of 1,000 users who signed up in week 1, 400 return in week 2 (40%), 300 in week 4 (30%), and the curve flattens near 25% - that 25% plateau is the product's retained core.
In practice: We teach teams to run cohort analyses on their own data in the Amplitude training practice.
Cohort Analysis: common questions
What is the difference between cohort analysis and segmentation?+
Segmentation groups users by an attribute (plan, country, device) at a point in time; cohort analysis groups them by when they started and follows each group forward. Cohorts answer 'is the product getting better?'; segments answer 'for whom?'
Get a free, scored audit of your Amplitude instance
Send us read-only access and get a scored findings report within 48 hours: taxonomy health, duplicate events, governance gaps, and the three fixes with the highest data-trust payoff. No commitment.
Request the free audit