I agree with you 100%. No options are ideal and statistically correct which sucks... a lot. As you mentioned there will always be selection bias and a black box

I agree with you 100%. No options are ideal and statistically correct which sucks... a lot. As you mentioned there will always be selection bias and a black box, just hopefully less, with Nielsen for example. Not many people in our space even understand these problems exist, so I think you already lead the pack in this regard. For the sales lift studies, the methodology seems less "iffy" since its one to one and measurable results. One can argue there's selection bias due to no cash data. I haven't seen any cheaper options to do this while minimizing error also. My client is in the same chicken-egg problem and the way we approached it is: Going on blind is worse than the risk of a bad study. It's completely up to us to do due diligence and ask the right questions to find holes in the vendor's methodology and minimize the risk. We felt that the data gained will be accurate directionally, albeit skewed. I've never had first hand experience with Adobe's attribution models but from what I gather, lack of data and similar to media mix modeling. IMO, advancements in this space will be battling privacy concerns.