I've been researching and working with this for some years now. There's a lot of stuff that can interfere with the quality/accuracy of the numbers (e.g. MAC randomization). I found out that there's also various ways to circumvent and improve upon those problems.