Names and aliases
interpurchase interval, natural demand rhythm
Formula
distribution(timestamp_of_transaction_n − timestamp_of_transaction_n−1) across demand participants in the micro-market, retaining censored intervals
Metric card
| Unit | days, reported as distribution median and percentiles |
|---|---|
| Data source | Completed transactions with a demand-participant identifier and timestamp, the micro-market definition, and account deduplication rules. |
| Frequency | By micro-market, using an observation window comfortably longer than the expected interval. |
| Segments | micro-market, category, geography, need type, cohort |
Interpretation errors
- Reporting an average number of purchases per month: it blends a small group of frequent buyers with a majority of infrequent ones and describes neither.
- Discarding participants with a single transaction: they carry the information that the interval exceeds the observation window, not that no interval exists.
- Transferring platform-measured frequency to real need frequency, when part of demand is satisfied elsewhere.
Example
Illustrative example: a micro-market shows a median interval of 94 days, a 25th percentile of 31 days, and a 75th of 210 days. The numbers describe no real platform.
Use
This is the anchor metric for choosing periods in every cohort calculation: the period length in retention curves and repeat rates should follow from the observed interval distribution rather than from the calendar. Compute it separately per micro-market and keep censored intervals visible. Publish the full distribution, because a median without percentiles hides the presence of two distinct demand populations inside one category.
Anti-example
The conclusion “a buyer makes 2.4 purchases per year” comes from dividing transactions by buyers. That average is identical for a market where everyone buys once every five months and one where 10% buy monthly and the rest buy once. The operational implications of those two markets are opposite.
Limitations
- The metric describes observed on-platform behaviour rather than participant need; unobserved purchases at competitors and offline are excluded.
- When the observation window is shorter than the typical interval, the distribution is systematically biased towards frequent buyers.
- Categories within one platform have incomparable natural rhythms, so a platform-level aggregate is almost always meaningless.
Related metrics
buyer-retention, buyer-repeat-rate, time-to-first-transaction, market-thickness
Context sources
These works provide two-sided-market context; the formula above remains an author-defined working definition.
- Geoffrey G. Parker, Marshall W. Van Alstyne. Two-Sided Network Effects: A Theory of Information Product Design (2005).
- Kong-Pin Chen, Yen-Chi Huang. A Search-Matching Model of the Buyer-Seller Platforms (2012).