Every platform reports itself generously. That is not a conspiracy, it is how last-touch works: whoever was standing closest to the purchase gets the sale written on their card.
A geo holdout is the cheapest honest answer available. Split comparable regions, switch a channel off in half of them, wait six weeks, compare. It is not perfect — seasonality, spillover and small samples all bite — but it is dramatically better than believing the dashboard.
In our experience, branded search is the channel that most often collapses under a holdout. It reports beautifully. Turn it off and revenue frequently does not move, because those people were coming anyway.
The second usual casualty is broad retargeting. Not all of it — but the segment made up of people who visited in the last three days and would have converted regardless is usually a large share of the reported return.
The uncomfortable part is that running these tests will occasionally prove that your agency's best-performing channel was never performing. We run them anyway, and we send you the result either way.