Killing an ad set before day 3 throws away the data you paid for

Killing an ad set before day 3 throws away the data you paid for

The in-platform CPA on day 1 is mostly noise — conversions report on a delay, and learning hasn't settled.

Reported vs final CPA by day (60 ad sets, tracked to day 7):
— Day 1 reported CPA: $52
— Day 3: $34
— Day 7 (true): $26

Killing on the day-1 number murders sets that finish profitable. 40% of 'losers' killed early would have beaten target by day 7.

Fix: set a no-touch window of 72h minimum (and through learning exit). Pre-commit a kill threshold in advance — e.g. 'pause if CPA >2× target after 50 conversions OR 3 days' — and ignore the panic in between.

Read: the chart on day 1 is a rumor; the chart on day 7 is the fact.