It's food for thought if you work in ad tech. Targeting based on tags/interests can be effective, but understanding the visual content of an ad and how it will be received by the user is a different problem that could result in better targeting if solved.
Suppose all ads were tagged for their visual content and their ad-hiding functionality looked for categories of content that you frequently hide. That data could be used for better targeting and passed all the way back to creative agencies to influence design decisions.
It is my belief that adtech companies' top priority is to make money. They do this by increasing clickthrough rate and cost per click. As long as they don't fall below some "trust" threshold (subjective emotional judgement from ill-informed customers pulling threads from obfuscated-at-best PR releases), they are golden.
From that perspective, why would they want to "better" target using proxy metrics for the ones you really care about? Focusing on the CTR[0], why use tags or interests when you can use clickthrough rate itself?
There are so many factors in what causes someone to click; it's a more effective strategy to treat all those factors as a black box, and optimize for the direct behavior you want to encourage.
Added bonus, those black box / supercomplex problems can be efficiently optimized by ML.
Suppose all ads were tagged for their visual content and their ad-hiding functionality looked for categories of content that you frequently hide. That data could be used for better targeting and passed all the way back to creative agencies to influence design decisions.
How is that not relevant to HN?