[lostinthefield]

It doesn't necessarily solve RoAS tracking either. If you only have information on some % of your purchasers* and you optimize against that small, self-selecting sample, you may entirely miss other groups of higher spenders who just don't want to be tracked.

It might help answer the question "which campaigns are performing better relative to each other" as long as we assume that no campaigns are more likely to affect opt-in rates than others. It may even be an improvement over up-front consent seeking, if the theory is that a better checkout UX with less upfront annoyances results in better opt-in rates later on. But it doesn't really answer the bigger question of "is this the best place to spend my marketing dollars" vs, say, investing in campaigns/platforms that may have greater reach but less transparent funnels (comparing a Google Ad to a Superbowl TV ad, for example).

If you use this in a vacuum without any statistical modeling thought, you end up optimizing against algorithmic limitations rather than real customers. At the end of the day you're drastically reducing your sample size, and not in a representative, random manner. Your opt-in users are behaviorally different from the majority of people, and it wouldn't be a good idea to make big decisions based on their actions alone. They are essentially a cheaper focus group, with all the pros and cons of such.

(*Probably a very low percentage, if the recent iOS tracking opt-in is any indication. One analysis says <15% of users opt-in: https://www.flurry.com/blog/ios-14-5-opt-in-rate-att-restric...)