The problem with stack ranking at places like microsoft was twofold. First small team managers were forced to rank their reports, and each manager had to get rid of the bottom 10% each year. If you had 20 employees you had to fire 2 employees each year. But statistics doesn't work like that, there were many microsoft teams with all productive employees. And there were probably teams where 30% of the group should have been moved/retrained/let go.

The second is that people respond to incentives. So a manager might make a bad hire because so they don't have to fire any of the people they like. The team might not help someone struggling because they need a sacrifice every year. And finally people might trade to barely avoid the bottom 10% (if you'll make this decision i want, i'll rank you higher). Then the people you are keeping are the low productivity, agile political operators. And a low productivity highly political worker is 10 times worse for productivity than the plain version of low productivity worker.

There doesn't seem to be any indication that Amazon is forcing this kind of decision on small teams. On a larger scale of 1,000 or 10,000 workers it makes sense to track if you are losing people that you don't care about losing or "un-regretted attrition". It's also good to have a process where a manager says "this person isn't doing well" and HR says "tell them what they need to improve and give them a few months" (or find them a different position). And the manager says "hey that worked" 2/3 of the time and then fires them when it doesn't. Otherwise you end up with sudden surprise firings or managers who never fire anyone.