You Should Know About Goodhart's Law
“When a measure becomes a target, it ceases to be a good measure” was a comment written by British economist Charles Goodhart in 1975, now known as Goodhart’s Law [1].
Almost 50 years on, I find this law to be more relevant than ever.
Out of the many workplace principles that I have come across, Goodhart’s Law is without a doubt my favorite one.
It reminds me to watch out for blind spots, which anecdotally, tend to lurk in the most critical areas of businesses and can inflict the most damage (more on this later).
As a software engineer who loves black-and-white data and metrics, Goodhart’s Law reminds me to beware of people’s plasticity in interpreting the spirit of targets.
“Any system will get gamed, watch out for unintended consequences” is my less elegant paraphrase of the adage.
An oft-quoted example of Goodhart’s Law took place years ago in Britain, whose government wanted its hospitals to reduce emergency wait times to within 4 hours.
The hospitals were able to meet the target by resorting to occasional gaming of the statistics.
One of the most egregious stories that emerged from this mandate was about patients who were made to wait in an ambulance rather than inside the hospital.
A patient was only admitted once the hospital was confident that the patient could be seen in 4 hours or less.
If you look closely, you will find that Goodhart’s Law plays out everywhere around you, perhaps even in your own workplace.
Let’s shift to recent events: this year’s layoffs at high-profile tech companies. This probably caught a lot of us by surprise, however, to some observers in the tech industry this was not unexpected at all.
They had long argued that many medium to large tech companies were bloated, and that corrective actions could be needed when the good times were gone.
You see, the promotion system in many tech companies is predominantly based on the concept of “impact radius”, which creates a perverse incentive for managers to continually expand the headcount under their management.
Motivated by their desire for promotion year after year, they kept coming up with justifications to form bigger and more numerous teams and sub-teams below them.
This, the observers added, created a “headcount bubble” that risked popping.
Ignoring Goodhart’s Law has resulted in a devastating outcome for the employees who lost their job.
So, how should you design your incentive system? Is it even possible to come up with one that is 100% manipulation-proof? Honestly, I don’t think so. However, I’ve had reasonable success with targets and metrics by taking these simple steps:
Put in place additional safeguard metrics. For example, if you give Sales an acquisition target of X new customers per month, you’ll want to monitor engagement level and churn rate to make sure that they are acquiring the right kind of customers.
Call it out upfront. When I presented our North Star to the team, I acknowledged that it was open to being gamed, which was why I asked them to always remember its spirit. Bringing it out to the open worked a lot better than I had expected. I think it’s a case of “You know that I know (about the possibility of gaming it) and I know that you know, so let’s not do it.”
Create a culture of honesty. Our North Star “moves up to the right” most of the time, but whenever one or a few metrics diverge and contradict it, there’s always at least one person who puts their hand up to look into it, to understand if the lift in our North Star is genuine. This public display of honesty discourages people from gaming our metrics.
Like I said, these are fairly simple steps. I don’t have a silver bullet I’m afraid. If you have tips to prevent perverse incentives, please let me know, I’d love to hear about them!
Footnote:
[1] This is technically a paraphrase. Charles Goodhart’s original wording was different.
Do you practice Basecamp’s Shape Up methodology at work? Check out our Shape Up Board for Jira plugin.
