Creator Media Buying With a Holdout on Every Scaling Call

No creative gets more budget here because Ads Manager said it worked. Scaling decisions are checked against a geo holdout that suppresses ten percent of eligible regions for at least twenty-one days, and the number that goes to your board is the one that survives that test. Platform-reported return is used for ranking creatives against each other and for nothing else. How the account gets structured: Few campaigns, wide ad sets, and automated buying only where the account can feed it. Below roughly fifty conversions a week at ad set level we do not run broad automation. The system cannot learn at that volume, it will report confidently regardless, and we would rather accept the lower ceiling of manual buying than hand you a number nobody can act on. Above that threshold, consolidation beats granularity and we will delete campaigns your team is attached to. Creative rotation: A creative comes out when three-day frequency crosses about 2.6 or cost per acquisition drifts more than a quarter off its first-week median, whichever lands first. We do not re-upload a winner to reset it. We do not run a fourth variant of a concept that has already lost twice to another concept. Rotation is scheduled against the production round so the next set of files exists before the current set fatigues, rather than three weeks afterwards while spend sits in a losing ad. Measurement: Conversions API in place, server events deduplicated against browser events, and event match quality watched at dataset level with seven as the floor before we trust an optimisation signal. Below that the platform is optimising towards a blurred picture of your buyers, and any creative test running on top of it is measuring the blur rather than the creative. What we count, by category: For gambling operators, cost per first-time depositor and thirty-day net revenue, with affiliate-attributed traffic excluded from test cells, because affiliate volume moving on its own schedule will wash out a holdout read entirely. For publishers, cost per paid subscription measured after the trial converts, never trial starts, which flatter every dashboard and distort every budget conversation that follows. For freight and logistics, conversion volume is too thin for any of this, so we weight cost per lead by CRM stage and accept a sixty-day reporting lag instead of pretending to a weekly read. Fees: Fixed monthly, never a percentage of spend. A percentage pays us to spend more, and we will regularly recommend spending less: pulling a channel, cutting a market, holding budget until the next production round lands. We are not prepared to hold both positions at once and we would rather not be asked to. We will not: Report last-click. Promise a cost per acquisition before seeing ninety days of your account. Run a test cell we cannot cleanly separate from an affiliate or PR spike. Extend a holdout past the point where it stops being informative in order to preserve the appearance of rigour. Buy in a market where the creative licence has expired, which causes more paused campaigns than anyone admits in a quarterly review. Who should look elsewhere: Advertisers under roughly forty thousand euro a month in creator-attributable media. The holdout will not reach a usable read and the fee is better spent on production. Teams who need a weekly return figure to defend budget internally: this method produces a firm answer about monthly and a lot of honest uncertainty in between. If that is politically unsurvivable inside your company, say so before contracting rather than in week six.