First-purchase conversion lag: D0 vs D3 attribution sensitivity
SKAN window 0–2 vs 3–7 and MMP Day-0 vs Day-3 purchase rates are different clocks. Illustrative sensitivity only.
SKAN and MMP windows disagree on first-buy timing. A D0 purchase rate is not interchangeable with a window-2 coarse conversion value.
Documented clocks
Apple’s Receiving postbacks in multiple conversion windows (source read 20 September 2026) states SKAdNetwork supports three conversion windows beginning at first launch: days 0–2, days 3–7, and days 8–35, with up to three postbacks when participants meet SKAN 4 conditions. Locking a conversion value early is optional via the lock-window API.
AppsFlyer Cohort treats Day 0 as the conversion day and lets you view on-day versus cumulative in-app event conversions. A cumulative D3 first-purchase rate includes D0–D3 buyers; an on-day D3 rate does not.
Illustrative sensitivity
Teaching cohort of 10,000 attributed iOS installs:
| Metric | Value | Notes |
|---|---|---|
| MMP unique first purchasers by end of D0 (cumulative) | 250 | Studio `af_purchase` first occurrence |
| MMP unique first purchasers by end of D3 (cumulative) | 410 | Same event |
| Share of D3 buyers missing from D0 | 39.0% | (410−250)/410 |
| SKAN window-1 (0–2) CV schema encodes “any purchase” | yes/no bit | Studio schema, not Apple revenue |
If UA optimises only to a D0 MMP purchase event while finance reads D3 cumulative payers, the lag alone moves the apparent conversion rate. Do not invent a “true” industry lag.
Facts vs limits
| Fact | Limit |
|---|---|
| SKAN windows 0–2 / 3–7 / 8–35 are documented | Mapping CV bits to first purchase is a studio schema |
| MMP Day 0 vs cumulative D3 are distinct views | This lab claims no genre D0→D3 lag average |
| Postbacks can be delayed after window close | Random delay ranges are documented separately by Apple |
