When to freeze a cohort for payback review (and when not to)

Moving freeze rules mid-argument rewrites history. Use documented data-freshness and path-freeze behaviour as the operator floor.
A payback argument that quietly changes which days are still “open” for late revenue is not a measurement debate — it is a definition rewrite. Freeze the cohort on a written rule before the review, and record when the freeze does not apply.
What the freshness docs actually freeze
AppsFlyer data freshness (source read 20 September 2026) states that for People-Based Attribution (PBA), during the first seven days after conversion PBA may identify additional events and change KPIs and media-source attribution retroactively; after seven days, the paths are frozen. Raw data includes a final_data field that is true when data is final. SKAN postbacks received on a given day are processed at end of day UTC and are available by 10:00 UTC the following day. Continuous (near-real-time) attribution updates with a typical 15–60 minute lag.
Operator freeze checklist
- Write the freeze horizon in the review brief (for example: D7 cash ROAS frozen at D7+2 UTC after last late SKAN window of interest).
- Do not freeze while continuous pipelines are still backfilling the same install day without a documented lag allowance.
- For PBA-style web paths, treat day 0–7 as mutable unless
final_datasays otherwise. - Never reopen a frozen cohort to absorb a creative win without logging a definition change.
Facts vs limits
| Fact | Limit |
|---|---|
| PBA documents a seven-day then freeze behaviour | Classic mobile cohort LTV freeze is not the same product surface |
| SKAN reporting has an explicit next-day UTC availability | Apple’s random postback delay is separate (see SKAN windows docs) |
| Continuous data still lags minutes | This brief does not invent a universal studio freeze SOP |


