Build the repeat-purchase cohort by the date of the first confirmed order, and compare only people with the same completed observation window. A recent buyer has not yet had the same thirty days to reorder as a buyer from last month.
Tutorial log
Conditional buyers A and B first purchased on 1 September, C on 18 September. On 20 September, no one has yet completed the thirty-day window. Even if A has already purchased again, the "repeat within 30 days" metric is still preliminary. Do not record C as a final non-repeater just because two days have passed.
For each anonymized buyer, store the date of the first purchase, the end date of the window, and the fact of the next qualifying purchase within it. Define a qualifying order in advance: confirmed, not a technical duplicate, with clear treatment of cancellations and refunds. One person with multiple messages does not become multiple buyers.
How to count a completed group
Suppose 20 buyers have fully completed the selected 30 days and 5 placed a repeat order. The share is 5 / 20 = 25%. This is only a conditional example. New 8 buyers with an incomplete window are shown on a separate line and do not change the denominator of the completed group.
The number of repeat buyers and the number of repeat orders are different metrics. One person with three orders counts as one repeat buyer. For money, sum confirmed transactions separately and do not count a wallet top-up as a repeat purchase of a service.
Comparison limitations
If buyer identity is not established across devices, do not merge records by guesswork. State coverage and unknown links. Changes in assortment, fulfillment time, and traffic sources can affect repeat purchases; a difference between cohorts does not prove the effect of a single article.
Diagram: 1 — first purchase; 2 — same window; 3 — completed group only. Tagging tools help distinguish external entry points, but they do not replace buyer identification. Engagement services do not guarantee repeat orders.
Sources and boundaries
Metrika: tag processing. Checked on 19 September 2026. Examples and schemes were prepared by the editorial team; these are not experiment results and not a promise by the platform.
View the terms of the separate service. The form below relates only to the specified activity, and not to setup, research, or editorial placement.