The platform's incentive is to credit itself generously. The business's incentive is to know what actually moved.
A geo-holdout test answers the question attribution cannot. Run paid media in one set of geographies (test markets), pause it in another (control markets). The difference in conversions across the test period is the incremental contribution of the spend. Whatever the platform was crediting itself, this is the actual upper bound.
What the gap usually looks like.
Across the test programs we run, platform-reported conversions typically overstate incremental contribution by 25-60%. The exact number varies by category, brand awareness, and spend tier, but the direction is consistent. If you do not test, you are scaling spend off a credited number that has a known inflation problem.
The three measurements to keep.
- Incremental conversions. Test market actuals minus what the control would predict.
- True CAC. Ad spend divided by incremental conversions. The number to scale against.
- Attribution inflation. The percentage the platform was overstating before the test.