> right, but again, those are very likely probabalistic, population-level models that make assumptions about how to attribute credit--does it all go to the first view/click? how likely is the first view/click really the first view/click? do you instead apportion credit across clicks/views? how? it's somewhat useful at a population level, but not at all at an individual level, especially not for the tradeoff in privacy, anonymity, and autonomy.

So the idea is that you run the experiment, and this can then help you understand where you should attribute value, as you know that the only difference between the two groups was the exposure to FB ads.

Now, to be fair, unless FB is most of your spend, you still have a bunch of problems, but they'll wash out equally across conditions (theoretically, at least) so the estimate should be good.