Run Google and Meta at the same time, pull each platform’s reported conversion revenue for the same period, add the two numbers together, and compare the total to what your backend actually recorded. The gap is almost always uncomfortable.
Platforms overclaim by design. View-through conversions, modeled conversions where user consent was never granted, and conversion windows that credit a click from three weeks ago all flow into the number they report to you. They all point the same direction: up.
The attribution problem in plain terms
Your backend under-attributes for the opposite reason. It typically only counts last-click, direct revenue, missing the assists that impression-based advertising contributed earlier in the buying journey. So you end up with two broken numbers: platform dashboards that claim too much and backend reports that give too little.
Neither tells you what actually happened.
What holdout testing measures instead
A holdout test splits your audience into two groups. One group sees your ads. The other does not. You run both groups for the same period, then compare actual revenue between them. The difference is what your advertising genuinely caused, not what a platform’s attribution model chose to credit.
According to the piece by Benjamin Wenner at MarTech, holdout tests are the only reliable method to measure real incremental revenue from advertising. Every other approach either borrows credibility from the platform selling you the ads or relies on models that have built-in incentives to show positive results.
The operator takeaway
If you’re spending real money on Google, Meta, TikTok, or any impression-based channel and trusting the dashboard numbers, you’re making budget decisions on data the seller produced. Holdout testing removes that conflict. It’s more work to set up, but it’s the only read on whether the spend is actually pulling its weight.
