That's a popular perception but it's not totally true. You need to be able to measure improvements you make, which might make a large category of possible improvements untenable from a perf point of view. But if you can show satisfaction scores or latency numbers or action-conpletion metrics or even just customer comments saying it solved their issue, that's plenty reason to attempt the change. The system encourages you not to work on things that "feel good" but instead things that are measurable. There's pros and cons to this as you can imagine.
That all said, I think the bigger problem is a willingness to put out things that don't meet a high standard. Ex: the new web GMail launched while slower than the older one. This was done to get new features out to users faster, but it doesn't feel that great from a product excellence POV. There's no Steve Jobs figure telling us the Pixel 4 doesn't look great and as a consequence we should go back to the drawing board. It's more like there's a set of market survey results and the phone was made to check those boxes.