Exactly this. What I used to do is listen to the user's proposal and go build that. 80% of the time it's not the correct thing to work on. I know spend much more time talking with users about their problems/pains, and then use metrics to figure out what they are trying to do. The combination of these two + a good set of eyes will point you in the right direction. Very very hard to do, but it's at least a leap better from my previous approach.