There is a key component to all feedback mechanisms for people that you may want to consider for machines: it is up to the person to act on the feedback (or not).

A book I recommend to people is Thank You for the Feedback. I'd say a third of the book is how to give great feedback (could be valuable in this case) and two-thirds is about how you decide what to do with feedback.

Are there cases that systems should not accept the feedback, say in the case of known bias/fairness? Or are there moments that the feedback isn't realistic based on how the system works (e.g. churn could have been avoided with a lower price but the business won't allow that discounting)?