What Is Cohort Analysis for Ecommerce?
Cohort analysis is the practice of segmenting customers into groups (cohorts) based on shared characteristics or events within a defined time period, then tracking their behavior and metrics over time to isolate the impact of specific actions or conditions.
Why Cohort Analysis Matters in Ecommerce
Aggregate metrics hide the truth. A 40% repeat purchase rate tells you nothing about whether customers acquired in January behave differently from those acquired in June, or whether a new onboarding email sequence actually moved the needle. Cohort analysis forces you to ask: which groups of customers are actually valuable, and when did they become valuable?
For DTC operators, cohort analysis answers three critical questions: (1) Are newer customers less loyal than older ones? (2) Did that product change, pricing shift, or marketing campaign actually improve customer behavior? (3) Which acquisition channel or time period produced the highest lifetime value? Without cohort-level visibility, you're flying blind on retention, LTV trends, and the true ROI of operational changes.
Core Cohort Types in Ecommerce
Cohorts are defined by the event or attribute that groups them. The most common structures in ecommerce are time-based and behavioral.
Time-based cohorts segment customers by when they first purchased or signed up. A monthly cohort groups all customers who made their first purchase in January, another for February, and so on. This reveals whether acquisition quality or product-market fit has improved month-over-month. A weekly cohort offers finer granularity, useful for testing rapid changes. Quarterly cohorts smooth noise and are better for long-term trend spotting.
Behavioral cohorts group customers by what they did, not when. Examples: customers who clicked a specific email campaign, customers who purchased a particular product, customers who spent more than $100 on first order, or customers who came from a specific traffic source (paid search vs. organic). Behavioral cohorts let you isolate the impact of a single variable - did the new email sequence improve repeat purchase rate? Compare the repeat rate of the cohort that received it against the cohort that didn't.
Acquisition channel cohorts are a hybrid: they're time-based (first purchase date) but also behavioral (source = Instagram ads, Google organic, etc.). This is critical for DTC operators because channel quality degrades over time, and you need to know if that's a channel problem or a broader retention issue.
Reading a Cohort Retention Table
A cohort retention table is the standard format. Rows are cohorts (e.g., acquisition month), columns are time periods after the cohort event (week 0, week 1, week 2, etc.), and cells show the percentage of that cohort still active or purchasing.
Example: The January 2024 cohort had 100% of customers in week 0 (by definition - they just purchased). In week 4, 28% made another purchase. In week 12, 12% had purchased again. This tells you the January cohort's repeat purchase curve. Compare it to the February cohort: if February shows 32% in week 4 and 15% in week 12, February customers are stickier. That signals either better product-market fit, a better acquisition audience, or an operational improvement that happened between January and February.
The shape of the curve matters. A steep drop-off in weeks 0-4 followed by a plateau suggests most repeat customers come back quickly; those who don't, won't. A gradual decline suggests a longer consideration cycle. A curve that improves over time (later cohorts outperform earlier ones) signals positive momentum - either in product, marketing, or fulfillment.
Cohort Analysis for LTV and Unit Economics
Retention is one metric; revenue is another. A cohort revenue table tracks total spend per customer by cohort and time period. This is where LTV lives.
Calculate average revenue per user (ARPU) by cohort: sum all revenue from the January cohort in weeks 0-52, divide by cohort size. Compare to February ARPU. If January ARPU is $85 and February is $92, February customers are higher value - either they spend more per order or they purchase more frequently. If January ARPU is declining month-over-month, you have a retention or monetization problem.
Cohort LTV analysis also reveals the impact of price changes, product mix shifts, or upsell campaigns. If you raised prices in March, compare the March cohort's ARPU to February's. If you launched a new bundle in April, the April cohort should show higher week-4 ARPU if the bundle is working. Cohort analysis isolates the variable.
Practical Setup: What to Track
Start with a single cohort dimension: acquisition month. Segment your customer base into monthly cohorts based on first purchase date. Track repeat purchase rate (or repeat customer percentage) at week 4, week 12, week 24, and week 52. This is your baseline retention curve.
Next, layer in a second dimension: acquisition channel. Create separate cohort tables for paid search, email list, organic, and paid social. Do paid search customers have higher week-4 repeat rates than organic? If organic is outperforming, that's a signal to shift budget or improve paid targeting. If paid search is declining over time, the channel may be saturating.
Add a behavioral dimension: first order value. Segment customers into three cohorts: first order under $50, $50-$150, over $150. Higher AOV customers almost always have better retention. But if that gap is narrowing (lower AOV cohorts improving faster), it signals better product-market fit or improved onboarding.
Track these cohorts monthly. Set a cadence: every month, pull the previous 12 months of cohorts and update the retention and revenue tables. Look for trends. Are newer cohorts outperforming older ones? Are specific channels degrading? Did a product launch or marketing change move the needle?
Common Pitfalls and Fixes
Confusing correlation with causation is the biggest trap. If the March cohort has higher retention than February, it's tempting to credit the new email sequence launched in March. But March might have had better weather, lower competition, or a viral moment. To isolate causation, run a test: segment March into two sub-cohorts - those who received the new email sequence and those who didn't (or received the old one). Compare retention between the two. That's a true test.
Ignoring seasonality will mislead you. December cohorts always look different from January cohorts because of holiday shopping behavior and gift-giving. Compare December to December, January to January. If you're comparing December 2023 to January 2024 and seeing a retention drop, that's expected. Compare December 2023 to December 2022 to spot real trends.
Cohort size matters. A cohort with 50 customers is noisier than one with 5,000. If you're segmenting by channel and a channel only drives 30 customers per month, your cohort retention rates will swing wildly. Combine smaller channels or accept higher variance.
Survivor bias skews results if you're only tracking customers who made a second purchase. Always calculate repeat rate as (customers who repurchased) / (total cohort size), not as average spend among repurchasers. The latter ignores the customers who churned.
Cohort Analysis in Action: A Scenario
An apparel DTC operator notices overall repeat purchase rate is flat at 30%. Cohort analysis reveals the culprit: January through March cohorts have 32% repeat rate, but April onward drops to 28%. Something changed in April. The operator checks: product quality issues? Fulfillment speed? Acquisition audience shift? Turns out, a paid social campaign launched in April targeting a lower-income demographic. Those customers have lower repeat rates. The operator either refines targeting or accepts lower LTV from that channel and adjusts CAC expectations.
Another example: a supplement brand sees week-4 repeat rate improving from 22% (January cohort) to 26% (June cohort). That's a 18% improvement. The operator credits a new onboarding email sequence launched in April. But cohort analysis of acquisition channel shows the trend is only visible in organic traffic; paid search cohorts are flat. The operator realizes organic customers are higher quality, not that the email sequence worked. The lesson: test the email sequence on a paid cohort to isolate its impact.
A third operator uses cohort LTV analysis to justify a price increase. The January cohort (pre-increase) had $120 average LTV. The February cohort (post-increase) has $135 LTV. Looks good - but week-4 repeat rate dropped from 30% to 26%. Customers are spending more per order but purchasing less frequently. The operator decides the price increase is sustainable but monitors closely for further churn.
FAQ
How long should I track a cohort?
Track for at least 12 months to capture seasonal patterns and full customer lifecycle. For high-frequency repurchase categories (e.g., consumables), 26 weeks is often sufficient. For lower-frequency categories (e.g., furniture), 24 months is better. The rule: track until the repeat purchase rate plateaus - when additional weeks show no new purchases from the cohort.
Should I use daily, weekly, or monthly cohorts?
Start with monthly. It's granular enough to spot trends but not so granular that noise dominates. Use weekly cohorts only if you're testing rapid changes (e.g., a new email sequence every week) and have high traffic. Use daily cohorts only for very high-volume businesses or when debugging a specific incident. Quarterly cohorts are useful for long-term trend spotting but hide month-to-month variation.
What if my repeat purchase rate is too low to see patterns?
If repeat rate is under 10%, cohort analysis is still valuable but noisier. Expand your time window (track to week 52 instead of week 12) and increase cohort size (use quarterly instead of monthly). Alternatively, define 'repeat' more broadly: any engagement (email open, site visit, cart add) instead of just purchase. This reveals whether customers are still interested even if they're not buying.
Can I use cohort analysis for one-time purchase categories?
Yes, but reframe the metric. Instead of repeat purchase rate, track customer lifetime value, referral rate, or review/NPS score. Cohort analysis still reveals whether newer customers are more or less satisfied, whether acquisition channel quality is changing, or whether a product improvement increased customer satisfaction. The principle is the same: segment, track behavior over time, spot trends.
FAQ
How long should I track a cohort?
Track for at least 12 months to capture seasonal patterns and full customer lifecycle. For high-frequency repurchase categories (e.g., consumables), 26 weeks is often sufficient. For lower-frequency categories (e.g., furniture), 24 months is better. The rule: track until the repeat purchase rate plateaus - when additional weeks show no new purchases from the cohort.
Should I use daily, weekly, or monthly cohorts?
Start with monthly. It's granular enough to spot trends but not so granular that noise dominates. Use weekly cohorts only if you're testing rapid changes (e.g., a new email sequence every week) and have high traffic. Use daily cohorts only for very high-volume businesses or when debugging a specific incident. Quarterly cohorts are useful for long-term trend spotting but hide month-to-month variation.
What if my repeat purchase rate is too low to see patterns?
If repeat rate is under 10%, cohort analysis is still valuable but noisier. Expand your time window (track to week 52 instead of week 12) and increase cohort size (use quarterly instead of monthly). Alternatively, define 'repeat' more broadly: any engagement (email open, site visit, cart add) instead of just purchase. This reveals whether customers are still interested even if they're not buying.
Can I use cohort analysis for one-time purchase categories?
Yes, but reframe the metric. Instead of repeat purchase rate, track customer lifetime value, referral rate, or review/NPS score. Cohort analysis still reveals whether newer customers are more or less satisfied, whether acquisition channel quality is changing, or whether a product improvement increased customer satisfaction. The principle is the same: segment, track behavior over time, spot trends.