Not one feature.

All analytics systems in general.

Whether it's implementing some tracking script, or building a custom backend for it.

So called "growth hackers" will hate me for this, but I find the results from analytics tools absolutely useless.

I don't subscribe to this whole "data driven" way of doing things, because when you dig down, the data is almost always wrong.

We removed a table view in favor of a tile overview because the majority seemed to use it. Small detail: The tiles were default (bias!), and the table didn't render well on mobile, but when speaking to users they told us they actually liked the table better — we just had to fix it.

Nokia almost went under because of this. Their analytics tools showed them that people loved solid dependable feature phones and hated the slow as fuck smartphones with bad touchscreens — the reality was that people hated details about smartphones, but loved the concept.

Analytics are biased.

They tell dangerous lies.

Did you really have zero Android/Firefox users, or do those users use blocking extensions?

Did people really like page B, or was A's design better except for the incessant crashing?

If a feature increased signups, did you also look at churn? Did you just create a bait marketing campaign with a sudden peak which scares away loyal customers?

The opinions and feelings of users are not objective and easily classifiable, they're fuzzy and detailed with lots of asterisks.

Invite 10 random people to use your product in exchange for a gift coupon, and film them interacting & commenting on usability.

I promise you, those ten people will provide better data than your JS snippet can drag out of a million users.

This talk is pretty great, go watch it:
https://go.ted.com/CyNo