Thread · 2 tweets · 9 Feb 2020

That's a big, unproven "if". If walled behavioural performs that well, why do the ad companies with the largest gardens spend so much effort acquiring 3P data, often at high legal and reputational risk? I don't think that anyone knows if FB's effectiveness comes from 1P or 3P.

Having said that I'm not opposed to client-side targeting, but building it on machine learning strikes me as a great way to make both fraud and discrimination simultaneously easier to implement and harder to audit, in exchange for unproven value.