Goodhart's Law

Goodhart’s Law happens when a measure becomes a target. Once the measure becomes the target, the measure doesn’t become useless; it becomes dangerous. The danger comes from trying to meet the target at any cost, gaming the system and resulting in potential future damage.

A more elaborated way to describe this phenomenon is Surrogation:
Surrogation is a psychological phenomenon in which the measure(s) of a construct of interest evolve to replace the construct itself. Research on performance measurement in management accounting has identified Surrogation as “the tendency for managers to lose sight of the strategic construct(s) the [performance] measures are intended to represent, and subsequently act as though the measures are the constructs of interest” (emphasis in original).
An everyday example of Surrogation is a manager tasked with increasing customer satisfaction who begins to believe that the customer satisfaction survey score is customer satisfaction.
Nobel prize winner Daniel Kahneman and Yale professor Shane Frederick postulate that three conditions are necessary to produce Surrogation:

• The objective or strategy is relatively abstract.
• The metric of the strategy is concrete and conspicuous.
• The employee accepts, at least subconsciously, the substitution of the metric for the strategy.

How to avoid Surrogation

• Get the people responsible for implementing a strategy to help formulate it.
• Loosen the link between metrics and incentives.
• Use multiple metrics.

An example of a good combination of metrics in technology teams are the Four Key Metrics. They balance each other in a way that if one metric is gamed, it will harm the other ones. For example, deployment frequency gets balanced by change fail percentage, if you deploy many new releases without taking into account adequate testing, you may increase the change failure percentage. If you don’t deploy because you don’t want to increase the change failure percentage, well, you have 0 deployment frequency.

We can get another example from banks: Imagine you are a bank, and you want a massive loan book because that’s the metric you are measuring. You only need to lend money to everyone. I don’t think I need to mention how bad for the bank that would be. A better metric would be a composite of loan book size with the credit risk of that book. With those two metrics, you can protect the future of the bank by not optimising growth with adverse effects.

Further reading: Don’t Let Metrics Undermine Your Business

Obsidian Links
[[resources-tools]]

Obsidian Tags
#mental-models