This is one of those fun things that you might shoot yourself into the foot by trying to estimate.
Piss a user off once and you may have a much harder chance of bringing them back. Sure, you can use the fact you drove away user X in your model for how much to annoy user Y, but that doesn't un-piss-off user X. Sooner or later, there might be a significant number of users X!
For instance, I've specifically avoided "real TV" cable packages even when I can get them for cheaper than streaming TV because my use case (sports playoffs and such) is on and off, and Comcast et al have successfully trained me that it's extremely unpleasant to cancel wired cable TV, so I'd rather pay more for something less obnoxious.
It's easy to be both "data driven" and foolishly short-sighted.