Not GP, but I can humbly share mine.

I think it's because a lot of analytics are data points oriented instead of being workflow oriented. So you can see that feature A is not being used a lot, but it's very important in a particular flow. Feature B may be used a lot, but it can be only important for a particular class of users while very detrimental mentally for another class.

People use software for a needs, but rarely I've seen a need being highlighted when interpreting analytics data.