Well, the alternative is working for days / weeks on a feature that the end-user may not even use (is the message I'm getting from this article). Being able to create and alter a feature quickly is important in an agile environment.
What you want to know is not just whether people click the button , but whether having a certain feature available changes the way that people actually use your application.
What if we graph this thing over a month? Hmmm, sort of always looks like a straight line for all of our test client datasets, and you can't really gain any insight from it. Let's try something different. etc etc.