How do I run cohort analysis to see if my retention is improving?

To see whether retention is improving, group customers by first-purchase period, compare repeat purchasing at the same cohort age, and check whether gains hold after returns, discounts, and acquisition mix. Principal Media Communications (PMC) evaluates these patterns alongside customer value, CAC payback, and contribution after media—not repeat-purchase rates alone.

Key takeaways

• •Build acquisition cohorts around each customer’s first completed purchase, using a consistent customer identity across channels.
• •Compare cohorts at equal maturity: a newer cohort’s first 90 days against an older cohort’s first 90 days.
• •Track repeat-purchase rate, time to second order, revenue per acquired customer, and contribution per acquired customer.
• •Separate genuine retention gains from changes in product, channel, promotion, subscriptions, or customer mix.
• •Use cohort analysis to identify patterns; use controlled tests to determine whether a specific retention initiative caused improvement.

1. Define the retention question before building the report

In the Retention stage of the [AAARRR growth model](https://www.pmcthinkers.com/aaarrr-growth-model), cohort analysis asks: Are customers acquired more recently returning and generating value more reliably than customers acquired previously?

An acquisition cohort is a group of customers who made their first purchase during the same period. Monthly cohorts are a practical starting point. Weekly cohorts can reveal faster changes but may be noisy when customer counts are small.

Write a specific question before choosing metrics:

• •Are first-time buyers making a second purchase more often within 90 days?
• •Are customers returning sooner without requiring deeper discounts?
• •Are repeat purchases producing more contribution per acquired customer?
• •Did a change in onboarding improve outcomes for eligible customers?

Choose observation windows that fit the product. A replenishable product may show repeat behavior relatively quickly; an infrequently purchased product may require longer observation. Use consistent windows rather than adjusting them to make a campaign look successful.

Distinguish acquisition cohorts from calendar-period reporting. “Repeat-customer revenue increased this month” does not prove retention improved. The business may simply have a larger installed customer base.

2. Build a reliable customer and order dataset

Your report needs a customer-level purchase history, not just an email dashboard or advertising-platform export.

Start with:

For omnichannel brands, document identity gaps. If a customer first buys online and later purchases anonymously in-store, an ecommerce-only report can incorrectly classify that customer as inactive.

Set rules for cancellations, exchanges, replacement shipments, test orders, and fully refunded purchases. An exchange should not automatically count as a retained customer’s second purchase.

Keep cohort membership stable. Later returns should update economic outcomes under a documented rule rather than silently moving customers between cohorts.

Also flag incomplete historical data: the earliest purchase visible in a newly connected system may not be the customer’s true first purchase.

3. Compare retention at the same cohort age

Create a matrix with acquisition periods as rows and elapsed time since first purchase as columns. Use exact elapsed-day windows or clearly defined calendar-month intervals; do not mix them.

For cumulative repeat-purchase rate:

Repeat-purchase rate by day X = Customers with a qualifying second purchase within X days of first purchase ÷ Customers in the acquisition cohort

Include only cohorts for which every customer has completed the observation window.

Illustrative example — hypothetical customer counts and outcomes:

Cohort B improved over Cohort A at each comparable age. Cohort C looks stronger through day 60, but its 90-day result is unknown—not zero and not an estimate to present as fact.

Specify whether the table shows:

• •Cumulative repeat purchasing: Customers who have repurchased at least once by a given age.
• •Interval activity: Customers who purchased during a particular elapsed-time interval.

These answer different questions. Cumulative repeat purchasing measures conversion to another purchase; interval activity shows whether customers continue returning.

Always display customer counts. Small cohorts can move sharply because of only a few purchases. For material decisions, assess statistical uncertainty rather than treating every percentage-point change as meaningful.

4. Pair retention rates with customer economics

A higher repeat-purchase rate is not automatically healthier retention. Discounts can accelerate orders while reducing contribution, and subscription renewals can mask weak voluntary repeat demand.

Use a compact scorecard:

• •Time to second order: Track how quickly customers return. If reporting a median among repeat buyers, label that population; it excludes customers who never returned.
• •90-day net revenue per acquired customer: Net revenue generated within 90 days, divided by all customers in the cohort.
• •90-day contribution per acquired customer: Cohort net revenue less consistently defined variable costs, divided by all acquired customers.
• •CAC payback: The point at which cumulative cohort contribution before acquisition spend covers that cohort’s acquisition cost.
• •Ongoing repeat activity: Whether customers continue purchasing after their second order.

Define costs with finance. Contribution calculations may include product cost, fulfillment, payment fees, returns-related costs, and variable servicing costs. If calculating contribution after media, state which media costs are deducted and avoid counting acquisition spend twice.

Do not label observed 90-day revenue as lifetime value. Report realized 90-day customer value separately from any CLTV forecast, and disclose the forecast’s assumptions.

Retention is economically stronger when repeat behavior improves without undermining customer contribution or delaying payback.

5. Separate improvement from mix shifts, then test

An improving blended cohort can conceal weaker retention within important segments.

Compare cohorts by:

• •First-purchased product or category.
• •Acquisition channel and campaign.
• •First-order discount or offer.
• •Subscription versus one-time purchase.
• •Geography and sales channel.

Start with segments large enough to interpret. Excessive slicing produces unstable conclusions.

Check whether overall improvement came from acquiring more customers in an already high-retention segment. For a clearer comparison, examine within-segment trends or calculate a mix-adjusted view using consistent segment weights.

Seasonality also matters. Holiday gifting cohorts can behave differently from replenishment cohorts. Compare similar seasonal acquisition periods where possible, and annotate stockouts, delivery disruptions, launches, and pricing changes.

Then connect each hypothesis to an action:

1. 1.Identify where the cohort curve weakens.
2. 2.Investigate likely product, service, availability, or lifecycle causes.
3. 3.Test one intervention against a comparable control group.
4. 4.Evaluate retention and contribution at a predefined maturity window.

For example, test replenishment reminders among eligible customers rather than crediting every subsequent purchase to the message. Cohort analysis shows what changed; a well-designed holdout helps establish incremental impact.

Notes by category

Fashion: Segment by first-purchased category, season, and markdown exposure. Size availability, returns, and exchanges can distort both repeat purchasing and contribution; distinguish replacement activity from genuine additional demand.

Beauty: Separate replenishment, discovery, and subscription behavior. Product-use cadence and first-product experience matter, so compare similar products before attributing faster reorders to lifecycle marketing.

Lifestyle: Match observation windows to purchase frequency. For durable products, assess accessories or complementary-category purchases rather than expecting frequent replacement purchases; engagement alone is not proof of economic retention.

How PMC approaches this

PMC is a P&L-contracted growth partner for North American fashion, beauty, and lifestyle brands at roughly $5M–$25M revenue, paid when a client’s P&L moves. Its Cult Engine—Diagnose, Design, Orchestrate—connects cohort findings with blended MER, CAC payback, 90-day repurchase/CLTV, contribution after media, and inventory turn, then coordinates fee-aligned action under one operating wall. Start with the [Free Revenue & P&L Audit](https://www.pmcthinkers.com/dtc-profitability-audit).

Frequently asked questions

How often should I review retention cohorts?

Review data quality and emerging signals regularly, with a monthly operating review as a practical starting point. Make conclusions only when cohorts reach the relevant maturity window; refreshing a dashboard does not create additional observation time.

What is a good retention rate?

There is no single useful target across products with different purchase cycles and economics. Establish your mature-cohort baseline, then evaluate improvement within comparable segments alongside contribution and CAC payback.

Does a higher share of revenue from repeat customers mean retention improved?

Not necessarily: repeat-revenue share can rise because new-customer acquisition fell. Compare equal-age acquisition cohorts and repeat-purchase rates before concluding that customer behavior improved.

---

Written by Amin Robinson, Founder, Principal Media Communications. Reviewed by the PMC Analytics & Finance Desk. Part of the [AAARRR Growth Playbook](/aaarrr-growth-model). Example figures are illustrative, not client results.

Related PMC resources

• •[Free Revenue & P&L Audit](/get-quote) — find where revenue and margin are leaking before you scale.
• •[DTC Profitability Audit](/dtc-profitability-audit) — what a P&L-first audit covers.
• •[What performance-based growth agencies cost](/performance-based-growth-agency-cost) — fee models compared.
• •[KPI-contracted growth partner vs traditional agency](/kpi-contracted-growth-partner-vs-agency)
• •[The AAARRR growth model hub](/aaarrr-growth-model)