Most of the job is measurement
Buying impressions on TikTok or Meta takes a few minutes. Working out which of them led to a sale takes much longer. It needs a pixel that fires on the right events, server-side events through the TikTok Events API or the Meta Conversions API on the pages we run, deduplication so one purchase is not counted twice, and UTM parameters that survive the checkout.
All of that is set up and tested before the first ad. After launch it is checked against the brand's own records, because a platform's conversion count is partly modeled, while an order in a store backend either exists or it doesn't.
Tests start small
Early on, the only question is which product angle, creative and audience convert at all, and a small budget answers it about as well as a large one. So tests start small and change one variable at a time. Each ad group or ad set has a stop-loss threshold set before launch, and one that spends double its target cost per result without converting is paused.
Most first tests fail. That is acceptable when they fail early and leave a written note on why.
Our fee depends on the result
For brand partners we usually fund the media and are paid per agreed result. When a campaign in our own ad account does not work, we carry the spend, and the partner pays only for the results actually delivered, which may be none. That makes us choosy about the work we take on. And with no retainer or hourly fee to fall back on, we gain nothing by recommending more spend than the numbers support.