Thursday, January 30, 2014

How to Fix a Stack Rank System: An Open Letter to the Next Microsoft CEO

With reports of a new CEO on the horizon I wanted to take some time as an x-msftie to give a foot soldier's perspective on fixing the performance review system - core to fostering great talent.

Recently, Microsoft did away with the dreaded stack rank – the system that forces all performance reviews to a curve with fixed ratios of low, adequate and high performers.  However, this may be overkill.

From my time in engineering and field teams it is clear that the biggest issue with the stack rank is each team is fitted to pretty much the same curve.  In the diagram below you can see that each team gets the same amount of rewards to play with - regardless of the team's performance as a whole.  This means that a rockstar team will have to give around the same number of reviews as a failing one. 



The reason for this is to create an organization wide standard that brings consistency to the review process across their 100k+ employees.  However, it creates perverse incentives where great performers will be paid more to contribute to weak teams than strong ones.  Great steadily revert to the mean as they compete to be mediocre +1.  Failing teams over reward the complacent best of the worst.

However, this isn’t the fault of the system – it’s just a failure of implementation.  Here’s how to get it right: 

1. Product P&L

When I was at Microsoft there were about 22 layers of management between a salesman and an engineer. The only person with true cross-functional P&L for those two individuals is the CEO (true fact).  This prevents any hard accountability from being enforced.   Products need to be re-structured into independent silos.   The organization needs to be restructured into GM-level accountabilities directly responsible for revenue.  While this will make collaboration more difficult, it does
prevent dead wood from accumulating and killing organizations.

2. Trickle-down stack.

A normal distribution can’t be assumed for all of your teams – simply put there will be good and bad performers.  So plan for it.  Each organization with P&L should be fitted to a curve (see below). 


Placement on this curve should dictate the skew of the distribution of reviews given within each team.  E.G. If you crushed it this year – everyone should be able to be rewarded.  If you failed, many more of you should get the axe.  This process will quickly filter people into superstar teams and remove deadwood quickly.

3. Intrepreneurship

As companies grow, they seem to deliberately remove incentives for employees to make revolutionary breakthroughs.   The intent is to avoid destabilizing an organization.  However, the result is always a stagnant top-heavy organization that is doomed to be overtaken by a younger rival (see disruptive innovation). In order to steadily create new products, individual employees needs to have the incentive and resources to break away from existing product teams to create their own.   When engineers only see a ~10% variation in compensation there isn't incentive to do 10x better work or be innovative.

No comments:

Post a Comment