zlacker

The tricky bit with ads is that someone is paying for a user to see it, so if the user never scrolls to it, or perhaps scrolls past too fast, you probably need to know so you don't bill for it. If the user taps on it you definitely need to know that, regardless of whether the ad link is working, but you can't redirect and slow down the ad link.

You need highly accurate data about user behaviour around the ads, and highly optimised display and linking.

I've built a similar system, albeit not for ads, for recommendations, based on viewing items in a feed. With a simple model, across web and app, all developed in-house, it was still hard and required a lot of care to get good ML signals and signals that people really understood.

Doing this across third party clients may be prohibitively difficult. I'd like to see attempts, but so far I've not seen any.

>>danpal+(OP)
Doesn't this problem go away if the site uses a CPC model?