Well, I just described exactly what it is and how it works, followed by a concrete example. So I'm not sure that's vague.

When the person taking the action is recording the action, and the information is kept private, then it's what I consider to be high quality information, even at very low levels of detail. Contrast this with the quality of information that ad networks get (it's lesser). Given that low-quality information the usual situation, I completely understand why you think that it takes a large amount of computing power to make useful predictions from it. I've found that the situation is greatly simplified when dealing with high quality information.