I see what you are saying: that there are too many confounding variables involved and it is not as simple as A implies B. I’d then argue the same thing about attribution then though: a company like Google or Facebook might be measuring many variables for a given user across however many surfaces they can possibly reach, but some set of offline variables that are invisible / unknowable from their vantage point might be the actual cause of a given conversion. If this is indeed the case, then measuring aggregate advertising metrics should be sufficient enough to inform ad spend in the long run.