Did you have any experience dealing with residential proxy networks? If so, can you give some examples of any mitigations to inbound requests from those connections?
Also, we refrained from calling it fraud, the term was invalid traffic. This also accounted for things like someone double tapping on an ad instead of once, and so on. There is a good report by Alexander Tuzhilin [1] commissioned by Google when they were sued for charging for invalid traffic that might give you some more ideas.
I was curious, when you say block traffic, would you send 400s or silently just not record the crucial events? It seems like maybe tipping off automation would cause them to readjust. Perhaps they are not that determined?
Things such as a single device ID sending hundreds of requests in a minute or blacklisted IPs would be just dropped silently, no point in even wasting bandwidth to show an ad. For cuts of traffic where the clicks etc were outside of predetermined ranges, it would again be blocked and dropped. In other cases, where fraud happened before we could block it, there was a flag called 'billable' which was set to False so the advertiser wasn't charged for it.
Who were the fraud operators affiliated with / what was their incentive? Isn’t it the publishers that would see value, or perhaps a competitor trying to disrupt marketing efforts?
Do you see opportunity for innovation in traffic validation? Have you seen much in the way of training ML for this purpose?
Fraud goes all the way from simple operations where actual people in countries like India or Bangladesh are physically clicking on ads to sophisticated operations where bots generate human like traffic which is mixed with genuine traffic to avoid detection.