Measure retention with cohort analysis

Total revenue can grow while the business is in trouble: if you keep acquiring new customers, growth hides the fact that they do not come back. Cohort analysis separates acquisition from retention by following groups of customers who started at the same time.

What a cohort is

A cohort is a group of customers who share a starting point, usually the month of their first purchase. The cohort table has one row per cohort and one column per month since the first purchase (month 0, month 1, month 2 and so on). Each cell shows the percentage of the cohort that made a purchase in that month. Month 0 is always 100%.

Reading the table

The example covers 18 monthly cohorts of the same 520 customers. Read it in three directions:

• Along a row: how one cohort behaves over time. The January 2023 cohort of 31 customers: 48% bought again in month 1, 42% in month 2, then retention declined to 13% by month 11.
• Down a column: how retention at the same age changes between cohorts. If month-1 retention rises for newer cohorts, your onboarding is improving.
• Along a diagonal: one calendar month across all cohorts. A diagonal that is unusually high or low points to a calendar event (a holiday season, a site outage, a price change) that affected everyone.

The average retention curve

On average, 44.8% of new customers buy again in the month after their first purchase, 35.8% in month 3 and 24.6% in month 6 (weighted by cohort size). The shape of the curve matters more than any single value:

• A curve that flattens out shows a core of loyal customers who keep buying; the business can grow by acquiring more customers like them.
• A curve that keeps falling towards zero means every customer eventually leaves; growth depends on endless acquisition.
• A smile (retention rising again later) means lapsed customers reactivate, often because of re-engagement campaigns or seasonal needs.

Comparing cohorts

The November 2023 cohort retained best in its first month (60.9%) and the January 2024 cohort worst (34.5%). Ask what was different about those periods. Customers acquired with a deep discount in a sale often retain worse than full-price customers; customers acquired before Christmas may be gift buyers. Cohorts by acquisition channel are even more revealing: if paid social customers retain at half the rate of search customers, their CAC must be judged against a lower lifetime value.

Retention, revenue and churn

Customer retention counts people. Revenue retention follows spend instead and can exceed 100% when remaining customers spend more over time (common in subscription businesses with upgrades). Churn is the complement of retention. In non-contractual businesses such as retail, you never observe churn directly (customers simply stop buying), which is why the next lesson uses a probabilistic model to estimate whether a customer is still active.

Practical choices

• Choose the period to match the purchase cycle: weekly cohorts for apps, monthly for most e-commerce, quarterly for durable goods.
• Ignore the last cells of recent cohorts: they are based on few months and few customers.
• Small cohorts (17 customers in June 2023) produce noisy percentages; look at patterns, not single cells.

Reporting

Cohort analysis of 520 customers acquired over 18 months showed that 44.8% made a repeat purchase in the month after their first order, 35.8% three months later and 24.6% after six months. First-month retention ranged from 34.5% (January 2024 cohort) to 60.9% (November 2023 cohort). The retention curve flattened after month 6, indicating a stable core of repeat customers.