More specific to insurance companies, the potential/plausible abuses to increase profit don't take much effort to ferret out. And to make matters worse the data they use will invariable flag a lot of false positives that you'll have little to no way of fixing. For example, my girlfriend regularly picks me up a pack of cigarettes on her way home from work. She usually uses her credit card instead of cash. Ok, now she's flagged in the system as being a chronic smoker and her rates get jacked up to holy hell. Except, of course, she's not actually a smoker and I'd be super skeptical about her ability to reasonably explain this to some random customer service rep over the telephone.
Data mining usage habits rarely result in lower prices for people making positive, less risky choices. Non-smokers will probably be giving a trivial token discount, but unlikely anything more.