Never solve a personnel problem with a policy

This is the inaugural axiomatic observation for a reason. Violation of this fundamental principle is frighteningly widespread, and its effects are broadly demoralizing.

The common form looks something like this: You’re an employee in an organization (company, university, whatever) and someone in said organization (call him S) makes a pretty bad mistake (M). Management’s next move should look something like this: S is taken to the woodshed, a proper willow switch is selected, and appropriate counsel regarding M is applied to S’s hindquarters. Other employees become aware that S now walks with a limp, and hence become aware of said woodshed experience, thus learning from the mistakes of others.

The problem seems to be that this course of action requires making an exception out of S in a way that may make management uncomfortable. It also requires deep individual thinking about the particular circumstances surrounding S’s behavior (M). I suppose in our modern litigious society it also exposes management to a lawsuit if they don’t treat everyone precisely the same way in the same circumstances. The problem is that the notion of identical circumstances is largely a myth. Every occurrence of M_i really has to be dealt with on an individual basis in order to actually be fair. That means time to discuss and deal subjectively, which makes many people nervous.

At this point, the easiest (read “most spineless”) way to manage is to craft a policy (P) which is then imposed on everyone in the organization, even though 99 out of 100 never had a problem with M. The real problem is when P has negative effects on innocent individuals, or (as is often the case) on the organization as a whole.

True story to back this up. Long long ago I was a manager in a company that will remain nameless to protect the guilty. I received a memo (which I still have in my files) pointing out that some managers were overspending their catering budget. Therefore… (drumroll please…) effective immediately, all catering will require a Vice President signature in order to be approved. I swear I’m not making this up.

What’s wrong with this picture? Gosh, where to start? How many managers were guilty? None that I personally knew. That’s not statistically pure empiricism, but I seriously doubt the problem was widespread. But even if it was, it’s not like they didn’t know who was blowing their budgets on donuts! Proper behavior would have been to systematically bring each offending manager (or both of them, or all three, whatever) in to the Vice President, who would select a suitable willow switch, etc.

Instead we now create the following scenario. A manager like me, who had never overspent any budget on anything ever, was punished in the following manner. First of all, do you know how much overhead there is in getting a VP signature on anything?! Could take weeks. So if I want donuts at my team meeting this Friday, I should have submitted a request in writing a week or so ago. But since the new policy was put into place, the VP’s desk has sprouted a signing pile of biblical proportion.

All that aside, we have to keep two things in mind. First, we’re talking about five bucks! Second, management is actually asserting with a straight face that the VP can make rationale, informed judgments about 30 instances of $5 worth of donuts on several dozen teams over multiple weeks better than the managers of those teams could have?! Still makes me crazy just thinking about it.

After the first couple attempts, I found that I’d wait a month, and then get 3 approvals in the same day, two of which were too late because the meetings had already happened. We began to figure out that with the influx of paperwork to the VP of Signature Bottlenecks, the stack would more or less sit until the overall numbers looked good, and then he would sign the entire stack (or the top half of it) in one mad flourish.

Ultimately my course was fairly straightforward. On Fridays, on my way to work, I would just go by the local donut joint, buy a couple dozen donuts myself, out of my own pocket, and bring them to team meeting. As a manager I felt that donuts were a positive thing in a meeting, and the amount of pain inflicted on me and my team by the system exceeded the $5 it cost me to just pony up and buy the refreshments myself.

But of course, it was the beginning of the end for the anonymous company. What does it suggest to anyone involved in the communal effort of constructing world-class and industry-leading software products when a few goofy excesses by an isolated pod of individuals leads to such a wildly inefficient and universally punitive policy for all? Suggests that there may be a leadership void somewhere between you and the top.

In machine learning, when an algorithm becomes too adapted to a particular training set, they call that overfitting the data. It’s bad, because while the algorithm smokes the test data, it ultimately fails (often badly) in the general case. It’s a fundamental result in machine learning and folks in that community are very cautious about it. Establishing a universal policy to correct isolated personnel problems is a classic example of overfitting the data. It’s like a conductor attaching radio tags to the ears of all orchestra members because the 2nd chair in the oboe section missed practice three times.

Leave a Reply

Your email address will not be published. Required fields are marked *