At one stage I was very into metrics and this was one of the more interesting aspects. As a very general rule, removing features basically amounted to the same

At one stage I was very into metrics and this was one of the more interesting aspects. As a very general rule, removing features basically amounted to the same effort as adding them (i.e. a feature that takes 5 weeks, generally took 5 weeks to remove). However, the payback in testing and regression was usually a no-brainer. You could make back your investment in removing something in 1-3 product cycles... Somewhat contrary to what you'd expect, this was often a better payback than adding features - of course you could take this to a variety of absurd conclusions... However, what it generally meant is that in most situations products will tend towards "feature bloat".