Well that's rather defeatist. Its also not correct. I've done a lot of work in data linking, which I consider basically the same thing, albeit not with advert

Well that's rather defeatist. Its also not correct. I've done a lot of work in data linking, which I consider basically the same thing, albeit not with advertising or ad fraud as the necessary subject matter. Now while its true that a lot of the things people do to lessen their fingerprint/signal is often not effective, its simply not true that such things are always a priori defeatable. Theoretically, you have to do some combination of plugging the leaks, increasing the noise, or dropping the signal. And if you do that then signal becomes harder and harder to extract. And its not even black and white like total win/lose. You could stop advertisers from finding out certain things, or change the relative value/cost of what's possible/profitable, or the degree of accuracy with which they can state certain things. Indeed, for instance, if you can get regular people's actions and behaviours to look more and more like these bots or fraudsters, that likely lessens the effectiveness of their machine learning techniques...