The sale happens where you cannot see it
Advertising that drives a purchase on a retailer's site or in a physical store produces no conversion an ad platform can record.
A brand can run a campaign that measurably increases sales and shows almost nothing in its advertising reports, because the transaction occurred on a platform it does not own.
That leads to a predictable error: cutting the campaigns that cannot be attributed and keeping the ones that can, which usually means over-weighting direct sales that are a small share of the business.
The available corrections are imperfect and worth using anyway. Sales data from retail partners, marketplace reporting, and geographic or time-based comparisons where a campaign ran unevenly.
Where direct sales exist alongside retail, they can serve as a signal rather than as the whole picture, provided nobody treats the directly attributable revenue as the total return.
Setting this expectation before spend starts is important, because the alternative is a quarterly review where a working campaign gets canceled for producing numbers the system cannot see.