This literally keeps me awake at night.
Statistic: it takes roughly 200 non-converting free trial downloads of my software to get one data point of feedback telling me why they decided not to buy.
If a user is already unhappy about dealing with you (safe to assume, since they're uninstalling your product. Fair to say?). It might be worth filtering whatever trolling you'd receive from such a feedback mechanism if it's that important to you.
As a personal interest of mine, what part of that 200 non converting trial downloads used the program up to expiry vs opened once?
Following the article's use case, the software could transmit the last dozen or so commands or actions performed by the user. In a simplistic scenario, how would you interpret it if 40% of your user's last commands are "Save"? Success, surely! How about "Undo"?
The art and magic, of course, are in figuring out valuable yet inexpensive metrics for your app, capturing that data when you have thousands or millions of users, and interpreting the meaning.
My feeling is that anything personal including user content and searches should be opt-in but opt out is ok for monitoring feature usage that can't be linked to an individual. Obviously online/web apps reveal at least this anyway. Does context web/mobile app/PC app make a difference?
Similarly it seems easy for metrics to mislead you into thinking the most important features for your app are the ones that get the most usage. It could easily be that some other features users would like far more but are currently hidden or poorly designed so that users are avoiding them