It sounds more like different data set that's gathered by observation. If you can keep the observer's bias to a minimum this method may tell you more about how your feature is received and used. It is also the most time consuming and costly approach and doesn't really scale.
Google: "Test 36 shades of blue and choose the one that performs best"
Apple: "Create beautiful product based on user experience and then convince you that you like it."
I don't think making all product decisions based on data is necessarily good, nor is designing in a vacuum.