So what happened ?
A few decision-makers quickly understood that to be “loved” by their ultra-protected employees, they’d better give them a 5-stars rating. On a ranking system going from 0 to 20, this means everyone was having a 20, except badly loved people (independant from actuel performance) which could possibly get a 19/20, or a minor annotation alongside the 20/20. Anyway, since it was impossible to fire bad elements, it became useless to make them even more angry with a bad note, which would “poison” the atmosphere for the months to come.

Obviously, managers which did not understood the strategy were quickly over-run, since the note tend to be “sticky” in such an administrative organisation, and had real effects/consequences on career and salaries opportunities.
The system quickly generalised, with everyone getting a 20, and occasionnal bad performers getting a 19. Attempting to give a 18 was bound to trigger a war with the local union representative.

It looks funny, but it was real. A full generation of administration workers have lived with such rules. You can guess that “bad performers” were completely hidden in this system, which suited them quite perfectly.

So the question is :
how to avoid “stack ranking” evaluation system, without leading into such system-wide collusion problems ?

It is worth noting that most large companies use “stack ranking” in some way, shape or form. It is widely perceived as a good practice, as it forces rigor in staff evaluation and development, and helps with “rating fairness” among other things. But like every “good practice”, it often fails to achieve its purpose because of poor execution.

There are many questions to consider when thinking about implementing “stack ranking”. Here are 2 specifically focused on the “bottom end” of the rating distribution.

1- Is my bottom quota aligned with my forced attrition expectations ?
If, as in the article, you expect 10% of low performers, you should be ready to ask 10% of your staff to leave. This % is on top of voluntary departures (medium- to high- performing employees that will leave you anyway). If you are not actually following through and letting people go, why are you telling these low performers that they are bad . In practice, except in high “forced attrition” industries, you are often much better off having a bottom quota <10%.

2- Am I enforcing the distribution at the right level ? This is basic statistics: if you expect 10% of your staff to be low performer on average , you need about 1,000 employees to be reasonably confident that the % of low performers will indeed be 10%. In a group of 10 people, there is a very high likelihood (read one chance out of two) that the right number of low performers is NOT 1, rather 0, 2 or more. This problem is more complicated than it looks, as it is quite hard to stack rank 1,000 individuals (most likely managed directly by over 100 managers), but getting that piece right is key to useful stack ranking

Back to MSFT, while I can easily see how stack ranking can go wrong, I believe that MSFT's lost decade's root cause has to be more than a poor performance evaluation system. I would not be surprised if the key issue was a poor decision making/accountability system instead.

The better alternative, as you highlighted, is to enable real-time feedback instead of stack ranking performance reviews. Real-time feedback gives employees the direction they need as they are working on a goal, not during a generic performance review when that feedback isn’t necessary anymore. This is how employees can grow and be better workers, which enables them to positively contribute to the organization.

As for stack ranking, it was big at GE in the days of Jack Welch. I’m sure that’s where Microsoft copied the idea. (Microsoft never being known for originality.) At GE, the practice has been discredited; now it is beginning to be discredited at MS.