Marketplace Retention: Measuring Who Actually Returns

A live marketplace often reports top-line growth while the founder cannot tell whether the same buyers actually return. If transaction volume increases solely because you acquire new users to replace those who immediately leave, it flags a measurement risk. However, missing repeat behavior is not always a product failure—especially for infrequent, lifecycle purchases. The decision to scale acquisition depends on retention, the normal purchase cycle, transaction contribution, supply availability, and cash reserves.

When Directorism designs measurement into custom marketplaces, we avoid platform-wide averages. We rely on defining a cohort—a fixed group whose first completed purchase occurred in a specified period—and tracking their repeat behavior to inform operational decisions.

Setting the Observation Rules

Before measuring who comes back, you need criteria for a return. As outlined in Amplitude's retention analysis documentation, retention requires tracking distinct users from a defined starting event to a defined return event.

Establish these baseline rules for your marketplace:

• The Event: Define exactly what constitutes a completed transaction. If a service is completed and later refunded, decide how that affects completion counts and contribution. Record the rule and apply it consistently across all cohorts. Exclude test data and resolve duplicates.
• The Identity: Use a stable buyer ID to count distinct people.
• The Timestamps: Record the exact first completion timestamp and return completion timestamps, standardized to a single timezone.
• The Horizon: Compare completed observation windows. Incomplete follow-up periods make newer cohorts look artificially worse.

The 90-Day Window

Break the user journey into elapsed intervals after their initial completion, without gaps: another completion up to 30 elapsed days later; more than 30 through 60 days; and more than 60 through 90 days.

Count each distinct buyer once per interval. Buyers returning in later windows do not need to have returned in earlier windows to be counted.

Building a Cohort Table

Start with a reproducible, plain-language method to track these distinct windows.

Consider a hypothetical cohort: 100 buyers who completed their first transaction in May 2026. You observe this cohort on September 1 or later, meaning all observation windows are complete. Every member has had a full 90 days to return.

• 32 make another completed booking in the first 30 elapsed days.
• 25 return in days 31-60.
• 18 return in days 61-90.

Using the original 100 buyers as the denominator for each interval, including the inactive people, your return rates for those specific intervals are 32%, 25%, and 18%. The latter two percentages are return rates for those specific intervals, not cumulative repeat-purchase rates.

Create a manual table to reconcile your counts:

An explicit, confirmed zero differs from missing data. If you have a newer cohort that lacks the full follow-up time, leave that cell unavailable or blank, not zero.

When interpreting these numbers, measure the time to second booking only among the repeaters, and always pair it with the fraction who repeat. Do not let non-returners disappear from your analysis.

The Economics of a Repeat Buyer

Gross Merchandise Value (GMV), platform revenue, and retained cohorts are distinct measures, as noted in a16z's marketplace metrics guide. Do not use total GMV as customer lifetime revenue. Instead of lifetime value, measure cumulative contribution over a stated observed horizon; any future lifetime estimate requires explicit assumptions.

Consider this hypothetical economics scenario:

• Completed Transaction: $200 USD
• Platform Commission: 20% ($40 USD revenue)
• Direct Costs: $12 USD (e.g., payment processing, direct support)
• Transaction Contribution: $28 USD (before acquisition and fixed overhead)
• Customer Acquisition Cost (CAC): $30 USD for one buyer

If a buyer books exactly once, that user yields a -$2 USD balance after that one $30 acquisition cost and before fixed overhead.

If that same buyer completes five bookings with the same economics, they yield $200 USD in revenue, minus $60 USD in direct costs, minus the single $30 USD acquisition cost, resulting in a $110 USD balance after that one acquisition cost and before fixed overhead.

Count CAC once, not per repeat. For deeper reading, review our guide on unit economics modeling and our unit economics calculator.

Supply Availability and Operational Context

Separate supply-side availability from buyer return. To measure supply retention, fix the original supplier group, define the return or availability event, and observe them over a complete window. Investigate marketplace friction through actual cancellations, unmatched requests, and support data rather than attributing cause from a buyer's absence alone. A lapse is a specific event, not a calibrated churn probability.

To manage this operational reality, use our Marketplace Operations Metrics Dashboard.

This Excel workbook acts as an operations context log containing:

• Read Me
• Weekly Dashboard: 12 rows (Week 1 to Week 12) tracking Active supply, Active demand, Searches, Requests, Matches, Completed transactions, Support tickets, and Disputes.
• Liquidity Metrics: A reference table of plain text ratios. (Learn more about marketplace liquidity metrics).
• Ops Queue: A log with Issue, User Side, Severity, Root Cause, Owner, Resolution, and Prevent Next Time?.

This shared download does not include formulas, charts, real-time updates, transaction imports, or a built-in cohort sheet. You must manually reconcile your data sources before entering the numbers. To track the retention cohorts discussed above, add a new "Cohorts" sheet manually to your locally saved copy of the workbook.

Define and reconcile a fully observed cohort before making changes to your acquisition spend.