"engage the user and see if you can get the feature to a common ground"

Yes, this is what I'm asking. In direct marketing for example, there are ways that success can be scientifically measured. I am wondering if any of you have used more methodical approaches to improving applications -- or if it's just guess and find out if it worked 2 months from now.

The reason for this is that I just don't always trust users to think deeply enough of the problem for them to know what they want.