Structure before spend
A working account has a cold layer finding new people, a warm layer retargeting everyone who engaged, and a conversion layer closing in-market buyers. Each needs its own creative, audience and budget line, because judging them against one blended number tells you nothing. We build that in week one, then keep consolidating — too many ad sets starve the algorithm of the data it needs to learn.
Tracking first, or nothing else counts
Since iOS 14 a browser-only pixel under-reports badly, and a campaign optimising on bad data gets confidently worse. We implement the Conversions API with server-side events, deduplicate against the pixel, pass hashed customer data for match quality, and verify in Events Manager before the budget goes up. This is the least glamorous and highest-return work in the account.
Creative is the targeting now
Meta's algorithm finds the buyer better than any interest stack — what it cannot do is rescue a weak ad. So the lever is creative volume: 15 to 20 new pieces a month, testing hooks and angles against each other rather than running three ads until they die. Most underperforming accounts are underfunding production, not media.
Scale without breaking what works
Doubling a budget overnight usually resets the learning phase and spikes cost per acquisition. We scale in measured increments, duplicate winners into fresh budget rather than editing live ad sets, and watch frequency so a warm audience is not shown the same ad eleven times. Reporting is cost per qualified lead or per order — ROAS alone hides the accounts quietly buying their own existing customers.