i feel you - that second-week drop-off is brutal and honestly, it's the single biggest efficiency killer in paid acquisition right now. been there, ran the same autopsy a dozen times.
Here's the thing: your budget is almost certainly fragmenting across too many ad sets or audiences, which kills the learning phase before Meta's algorithm can optimise. when you spread even £50/day across three countries and five ad sets, each gets a pathetic amount of data. No wonder creatives flatline by day 10.
The fix that worked for me:
- Pick one country - ideally your highest-LTV market or the one with lowest CPA in your historical data. For us that was UK (sometimes Germany). Run everything through one campaign with one ad set.
- Budget floor - give that single ad set at least 3x your target CPA per day. If your target CPA is £20, don't start below £60/day for that one country. Anything less starves the algorithm.
- Creative testing cadence - I use a three-phase structure: Day 1-3 (broad top‑of‑funnel hook tests), Day 4-7 (scale winners to 2x budget), Day 8-14 (refresh creative before fatigue hits). Without that schedule you're just hoping.
- Use a budget tracker - I built a simple sheet in Google Sheets tied to Meta's API via Supermetrics to flag when any ad set's ROAS dips below 0.8x target before week two hits.
Also worth checking your attribution window. If you're still using 7‑day click, switch to 1‑day click + 1‑day view for optimisation during the first two weeks. shorter windows help the algorithm learn faster instead of waiting for delayed conversions.
Start with one country, one ad set, one angle. Let it mature for a full two weeks before scaling or adding anything else. it's boring, but it stopped the "week two flatline" for us.
