−58% CAC, 4.1× creative volume

Meta CAC had doubled in six months. The brand was shipping 4–6 new ads a month, and audience fatigue set in by week two of every launch. The team read this as a quality problem. The ads must be weak, so hire a better creative agency and make prettier ones.

That read was wrong. Twelve months of ads were variations of the same founder-in-kitchen format the team happened to like. Meta’s algorithm needs diversity of hook, not diversity of background color, and no ad shipped with a hypothesis attached, so even the occasional winner taught the team nothing about what to make next. Volume and learning were the constraints. A more polished version of the same ad would have moved neither.

We audited twelve months of performance and mapped every winner into one of six hook categories: problem, authority, founder, result, comparison, social proof. Three of those categories had barely been produced against. That is where we went first.

Production got rebuilt around hypotheses. Every new ad started with a written brief naming the hook, the audience objection it answered, and the metric it was expected to move. Scripts went through a 24-hour feedback loop with the media buyer before shoot. Shoots ran weekly instead of monthly, and the ad-to-launch cycle settled at 12 days.

The biggest saving came from micro-testing. Before committing to full production, we ran three-day whisper tests, $200 against a rough mock-up, to validate the hook. About 40% of hooks did not clear the bar. Those never went to production. The 60% that did convert into finished ads hit at a much higher rate, because the weak ideas had already died at the $200 stage before real money touched them.

Four months in, creative output ran 4.1× higher, 5–8 ads a week against the old 4–6 a month, and Meta CAC had dropped 58%. The ads did not get prettier. Every ad tested something specific, the winners got budget, and the losers got killed cheap.