Users will give you the preferences that they think are important. Or are important at the point of that interview.
Analytics can give you a decent glimpse of revealed preferences, which may or may not be what you're after.
Whether or not this is a good thing, depends on a lot of subjectivity, sure. But suppose you run a porn site - if you asked most users what they wanted in porn (before they had seen any), they would probably say one thing. If you examine what kinds of videos people look at, you'll see another. (This theme, with actual data from pornhub, is explored at length in the book "Everybody Lies" by Seth Stephens-Davidowitz.)
Both routes (asking and instrumenting) have their uses.
Glad you brought up porn because I believe that if you optimized any site's features just based on engagement analytics, you'd end up with a porn site with elements of gambling!
I'm kidding of course, but the idea is that analytics tell you a part off the user story, but doesn't answer the deeper "why" questions. It certainly has a place in tech, but it's less than what it's currently afforded.
Bender was on to something. "X but with blackjack and hookers" is the next untapped market for disruptive startups. We've worn out "X, but on the internet" and "X, but blockchain".
Analytics can give you a decent glimpse of revealed preferences, which may or may not be what you're after.
Whether or not this is a good thing, depends on a lot of subjectivity, sure. But suppose you run a porn site - if you asked most users what they wanted in porn (before they had seen any), they would probably say one thing. If you examine what kinds of videos people look at, you'll see another. (This theme, with actual data from pornhub, is explored at length in the book "Everybody Lies" by Seth Stephens-Davidowitz.)
Both routes (asking and instrumenting) have their uses.