In this article
Repeat buyers can look healthy while repeat business becomes harder to find for most sellers. A marketplace that tracks only returning customers may miss that concentration until its available supply starts to shrink.
Read repeat use from both sides of the same transaction ledger. Then examine whether the same buyer-seller relationships recur and how much volume the largest participants receive. These views describe different properties of the market; none is a substitute for the others.
Keep the transaction as the common record
Each completed transaction needs a stable buyer, stable seller, completion time and value under a consistent GMV definition. If the buyer is a company with several employees, decide whether the customer unit is that company or an individual. Do the same for a seller business with multiple operators.
Amplitude’s account-level reporting supports grouping activity at an account or organization level. The relevant design principle is to choose the unit that owns the relationship. Grouping events technically does not decide which entity your marketplace should count.
Preserve user-level detail where it helps explain the work, but avoid counting a staff change as the departure of a buyer business. Likewise, two employee accounts should not create two retained sellers if they serve the same supplier organization.
Follow four buyers and four sellers
The table below is an invented marketplace with three equal, fully observed 30-day windows. Every row is one completed transaction. There are no refunds, additional transactions or new participants. B1–B4 identify buyers; S1–S4 identify sellers. Window 1 establishes the base cohort.
| Window | Buyer | Seller | GMV |
|---|---|---|---|
| 1 | B1 | S1 | $100 |
| 1 | B2 | S2 | $100 |
| 1 | B3 | S3 | $100 |
| 1 | B4 | S4 | $100 |
| 2 | B1 | S1 | $150 |
| 2 | B2 | S1 | $150 |
| 2 | B3 | S2 | $100 |
| 2 | B4 | S3 | $100 |
| 3 | B1 | S1 | $250 |
| 3 | B2 | S1 | $250 |
| 3 | B3 | S1 | $100 |
| 3 | B4 | S2 | $100 |
All four original buyers transact in both later windows. Buyer recurrence is therefore 4 / 4, or 100%, in each. GMV from these returning buyers grows from $500 in window 2 to $700 in window 3.
The seller result differs. Three of the four original sellers transact in window 2, for 75% recurrence. Only two transact in window 3, for 50%. A report celebrating returning-buyer GMV would miss that narrowing participation.
These are window-specific recurrence rates. A seller who skips one window and returns later still belongs to the original cohort. The empty window does not establish permanent departure.
Read pair recurrence with a named reference period
The base window contains four relationships: B1–S1, B2–S2, B3–S3 and B4–S4. Only B1–S1 appears again in window 2, and only that same base pair appears in window 3. Base-pair recurrence is 1 / 4, or 25%, in both later windows.
This tells a different story from buyer recurrence. The buyers continue using the marketplace, but most move to different sellers. Depending on the job, that can be ordinary: customers may need different expertise, locations or available times. For a product built around recurring relationships, it may deserve closer attention.
Always identify the pair denominator. Comparing window 3 with window 2 would count a different set of relationships from comparing both with window 1. Keep the original cohort view stable, and use a separate transition view when studying how relationships change from one window to the next.
Avoid interpreting pair recurrence as relationship satisfaction by itself. A buyer can return to the same seller because of limited alternatives, and can choose a new seller for a perfectly good reason. The ledger identifies the pattern; product research explains it.
Inspect where the volume goes
S1 receives $300 of window 2’s $500 GMV, or 60%. In window 3, S1 receives $600 of $700, or about 85.7%. Returning buyers are spending more while their activity concentrates on one supplier.
a16z’s marketplace metrics framework includes concentration on each side of a marketplace. Pair that view with recurrence so the team can distinguish broad participation from growth carried by a small number of accounts.
Concentration is not automatically unhealthy. A specialist may deserve more demand because it can handle the work. But the marketplace should understand its dependence: what happens to fulfillment if that supplier becomes unavailable, changes terms or reaches capacity? It should also understand whether other suitable sellers have a reasonable opportunity to receive demand.
Inspect buyer concentration separately. In the example, all four return, but equal participation does not imply equal spending. A marketplace can have broad seller participation and still depend heavily on one purchasing organization.
Ask what happened to the sellers with no repeat business
S4 receives no transactions after the base window. That does not reveal whether it left, stopped offering availability, received unsuitable requests or lost placement in search. S3’s absence in window 3 raises the same questions.
For each quiet seller, inspect active availability, exposure to relevant demand, responses and completed work. Those stages help distinguish a supply decision from a matching problem. If a seller had no suitable demand in its service area, a generic “come back” campaign is unlikely to resolve the issue.
Also check whether the marketplace changed how work is routed. A ranking change could concentrate demand; so could differences in supplier quality or capacity. The ledger alone cannot assign causality. Compare the relevant conditions and investigate before deciding to redistribute work.
Bring the views into one operating review
Use the same mature windows and base cohort for buyer recurrence, seller recurrence and base-pair recurrence. Show returning-buyer GMV and concentration beside them. Keep new entrants separate so acquisition does not conceal what happened to the original participants.
The fictional market would prompt a supply-side review despite its strong buyer numbers: fewer original sellers receive transactions, and one handles most of the volume. The next decision might concern availability, matching, capacity or the marketplace’s intended specialization. The numbers do not demand a single answer, but they prevent a one-sided celebration.
For the timing behind these comparisons, see reading a retention curve. When incorporating repeat use into the product growth model, carry both sides forward. A marketplace needs buyers who return and enough viable supply to serve the work they bring back.
Sources
Something we should clarify?
Suggest a correction


