Running a fixed 30-day lookalike refresh schedule sounds disciplined. Turns out, roughly 40% of those rebuilds fire when the audience is performing just fine — and every unnecessary rebuild resets Meta's learning phase.
The decay problem itself is sneaky. It doesn't spike your CAC overnight. What it actually looks like is a 15–25% cost creep over 6–8 weeks that reads, on the surface, like creative fatigue. I've burned entire A/B testing cycles on ad copy when the real culprit was a purchase-event seed built off a Black Friday cohort, still running months later in a normal-demand window. The lookalike model doesn't auto-update — it's a static snapshot. You're essentially targeting a behavioral profile of customers who no longer represent who's buying from you right now.
When I audited a batch of accounts, the average seed age at first review was 70–90 days. For brands doing consistent volume, I'd consider anything past 45 days a risk. The three early signals I watch: CPM climbing while CTR holds flat, CPC drifting up even when frequency is under 2.0, and seed age crossing that 45-day mark. Two out of three and I'm already pulling the audience to rebuild — no waiting for CAC to confirm it.
I ran two approaches head-to-head across two comparable brands (similar AOVs, similar ₩40M–₩60M monthly spend). Fixed 30-day calendar refresh vs. a signal-triggered system that only acts when decay indicators cross defined thresholds. The signal-triggered approach reduced wasted rebuilds from ~40% down to under 10%. The tradeoff: peak CAC overshoot was slightly worse (+22% vs. +18%) because you're waiting for confirmation before acting. The hybrid I landed on — signal thresholds as the primary trigger, hard 45-day cap as a ceiling — captures most of the precision without letting a slow decay run too long.
The counterintuitive finding that genuinely surprised me involves seed list size and decay rate. I'd assumed larger, higher-volume seeds would hold their quality longer. The data said the opposite.
I wrote up the full breakdown — including the exact decay mechanism behind that finding and the week-3 decision checklist I run on every new campaign — over on themedilog.com.
Top comments (0)