Cohort Analysis
Part of Retention cohort analysis
Building cohorts from a meaningful first purchase
Define the first qualifying purchase, resolve identity and order-status edge cases, and document a cohort start date that colleagues can reproduce.
Start a purchase cohort when a customer first makes the kind of completed purchase whose later repetition you want to study. A platform’s earliest order date is a useful candidate, but an unpaid, cancelled or fully returned order may be the wrong starting point for the business question. Write the eligibility rule before grouping customers by date.
Decide what “first” means
Ask whether the cohort starts at the first store order or the first purchase in a particular product group. Those are different populations. Someone who bought a gift two years ago and now buys a refill for the first time is an established store customer but a first-time refill buyer. Name the cohort accordingly.
Specify the order state and date used. An order-placed date gives an early timestamp; a completed or fulfilled date may better suit a question about receiving and using a product. If completion occurs later, preserve both dates and use one consistently for cohort assignment.
Decide in advance how to handle a later full refund or cancellation. Changing eligibility after seeing outcomes can silently change past cohort sizes.
For a category cohort, document which items qualify and whether a mixed basket qualifies. Identify qualifying items at line-item level where the data permits. An unrelated item in the same basket need not change the category entry date.
First Purchase Criteria: Store vs. Category Cohorts
- Store Customer CohortStarts at the customer’s first ever completed order with the business, regardless of product type.
- Category Cohort (e.g., Refill Products)Starts at the first completed purchase of a specific product group, even if the customer has bought other items before.
- Example ScenarioA customer bought a gift two years ago (counts in store cohort) but buys a refill for the first time today (new category cohort).
Make the first purchase comparable
Resolve three common ambiguities before grouping dates into months or weeks:
- Customer identity.Check whether online, in-store and guest orders can be joined to the same customer reliably. An unmatched older order can make a repeat customer look new. Keep uncertain matches separate rather than asserting they are first purchases.
- Order purpose.Flag subscriptions, replacements and gift services when they change the question. A subscription’s next scheduled delivery differs from a new decision to order; either may be measured, but the label must say which.
- Product opportunity.Group items with a plausible shared reason to return. A refill and a long-lasting appliance should not receive the same short follow-up expectation simply because both appear in one store.
Record the rule in a compact data dictionary: customer key, qualifying product set, order states, first-date field, refund treatment, channels covered and reporting cut-off. That makes later comparisons reproducible.
Cohort Definition Checklist: Ensuring a Meaningful First Purchase
- Confirm customer identity across channels (online, in-store, guest)Ensure all orders are linked to the correct customer profile using consistent identifiers such as email or phone number. Use separate cohorts for unmatched or uncertain matches.
- Define order state for cohort entry (e.g., completed, fulfilled, paid)Use 'completed' or 'fulfilled' date instead of 'placed' if measuring actual product receipt and usage. Maintain consistency across analysis.
- Specify qualifying products and basket rulesIdentify which items count as a first purchase in a category. A mixed basket is acceptable if at least one qualifying item is present; unrelated items do not invalidate entry.
- Document refund and cancellation policyDecide whether full refunds or cancellations after the first purchase disqualify the customer from the cohort. Apply this rule consistently before analysis begins.
- Distinguish between subscription renewals and new purchasesFlag subscription-based orders separately. A scheduled delivery should not be treated as a new decision to buy unless explicitly triggered by user action.
Audit boundary cases
Sample records immediately before and after a proposed cohort boundary. Include a cancelled order followed by a paid order, a full refund, a partial return, a guest checkout later linked to an account, a mixed basket and a customer with an earlier in-store purchase. Check whether the written rule assigns a consistent start date in each case.
A default customer cohort report may not establish a first completed category purchase under a retailer’s custom exclusions. Check the report definition against the data dictionary; calculate the cohort separately if the required entry rule differs.
Once entry dates are stable, group them into periods and record the eligible customer count at a stated reporting cut-off. Document later corrections to identity or order status so a revised cohort count can be explained.
Key Events in Cohort Entry Process
- Define eligibility ruleEstablish criteria for first purchase before any data processing.
- Audit boundary casesTest edge cases: cancelled order followed by paid order, partial return, guest checkout later linked to account.
- Apply rule consistentlyUse the same logic across all customers and time periods to avoid bias.
- Group start dates into periodsAggregate cohort entries into monthly or weekly buckets for analysis.
- Record cohort size at reporting cut-offDocument the final eligible count with a clear date and explanation of any corrections.
