MarketingFramework / modelModelAccessible

Cohort analysis

How can cohort analysis improve people, teams, or organisational effectiveness?

AccessibleStrategicTeam2 min read

Contents

Cohort analysis is a subset of behavioural analytics which allows you to study the behaviour of a group over time.

Cohort analysis follows groups that share a defined starting event or characteristic and compares their behaviour over time. Instead of blending every customer, employee or user into one average, it preserves the context of when or how each group entered the analysis. Data may come from e-commerce, applications, workforce systems or sales records.

When to use it

Use cohort analysis when an aggregate trend may be mixing groups with different exposure, maturity or behaviour. It is especially helpful for retention, adoption, repeat purchase, absenteeism and the effect of a product or service change.

Grouping customers by acquisition month, for example, can show whether apparent overall growth hides weakening retention among recent users. Grouping employees by start date or location can reveal patterns that an organisation-wide absenteeism average conceals.

The method helps answer:

Which customer cohorts become most valuable over time?

What entry characteristics or experiences distinguish the groups?

How do retention, purchase or engagement patterns change with cohort age?

Which employee cohorts share a material outcome and what exposure might explain it?

Origins

Cohort analysis originated in demography and epidemiology, where researchers followed groups defined by birth period, exposure or diagnosis to distinguish life-course patterns from events affecting everyone at the same calendar time. Norman B. Ryder’s mid-twentieth-century essay “The Cohort as a Concept in the Study of Social Change” became foundational in social research. Digital products and customer analytics later adapted the method to acquisition cohorts, retention curves and lifecycle behaviour.

What it is

A cohort is defined by a shared event or attribute and an eligibility rule. Acquisition cohorts begin in the same week or month; behavioural cohorts share an action; demographic or organisational cohorts share a characteristic. The outcome is then measured at comparable elapsed times—such as the seventh day, third month or first year—so older groups are not unfairly compared with newer ones.

This structure turns a large, mixed database into lifecycle patterns. It can reveal that a product change improved early activation but harmed long-term retention, or that one employee group experiences a problem only after a particular transition. Cohort evidence supports action when definitions are stable and plausible alternative explanations are examined.

How to use it

Use four steps:

Define the decision and question. State the stakeholder, behaviour, horizon and action the result could change. A useful question is specific enough to determine who enters the analysis and when observation begins.

Choose outcomes and attributes. Define retention, revenue, absence or another measure precisely. Select relevant properties such as acquisition date, first purchase, channel, gender or purchase level, while avoiding unnecessary sensitive attributes.

Construct comparable cohorts. Specify inclusion, start event, time window and exclusions before viewing results. Check sample sizes and use attribute-contribution analysis cautiously to identify differences worth investigating; association does not establish the reason for behaviour.

Analyse and visualise. Build a table or curve with cohorts as rows and elapsed time as columns. Compare levels and trajectories, add uncertainty where samples are small and test whether changes in season, channel, product mix or data collection explain the pattern.

Different questions require different cohort definitions. Document the logic in the software or query so the analysis can be reproduced and refreshed.

Practical example

Suppose a company wants to understand who buys its products. Total sales and broad demographics show composition, but they do not reveal how comparable groups develop after acquisition.

The company groups customers by acquisition month and then segments carefully by relevant demographics. It discovers that women aged 35–45 generate unusually high repeat sales across several cohorts. That pattern can justify research and a tailored marketing test, but not the assumption that age or gender caused the result. The team first checks channel, product and income differences, then tests whether a more relevant proposition increases incremental sales without stereotyping or excluding other groups.

Top practical tip

Anchor every group to the same starting event and compare it at the same elapsed age. Otherwise recent cohorts will appear worse simply because they have had less time to act.

Top pitfall

Historical cohort patterns are not forecasts by themselves. Product changes, channel mix, seasonality and self-selection can make a past relationship disappear or reverse.