For years the craft of paid social was audience building: stacking interests, layering lookalikes, slicing demographics. Meta has quietly taken most of that job back. Hand it a broad audience and a clear goal and its machine learning finds the buyers better than manual targeting usually can.
That does not make the marketer redundant. It moves the lever. The thing you still control, the thing that now decides whether a campaign flies, is the creative itself.
So we stop treating creative as the pretty bit added at the end and start treating it as the experiment. Each ad is a hypothesis about what will make someone stop, care and click. You run a lot of them, read the results honestly, and let the winners earn the budget.
No one reliably guesses the winner in advance, including us. What works is putting enough genuinely different angles into the auction that the algorithm has something to optimise toward, then doubling down on what proves itself. One brilliant ad fatigues. A system that keeps producing them does not.
The trap is judging creative on vanity engagement: likes, shares, clever comments. We judge it on whether it drives profitable action, and we give each test enough room to be fair before we call it. Then the tired creative is retired before it quietly bleeds budget, and the cycle starts again.
We will build a creative testing system that keeps finding winners.