Think About the Incentives

Defining and then measuring success can be one of the most reliable ways to ensure you achieve what you intend to. Done poorly, it is also one of the best ways to ensure you improve the chosen metric, regardless of your true goals.

Once you have taken a nebulous concept like success and made it concrete like profit, there is the potential for a gap between what’s desired and what’s measured. That gap can lead to surprisingly bad outcomes.

Some background

In my first game theory class, my professor implored us all to “Think about the incentives!” The math of game theory was challenging for me. The key thing we had to calculate was the Nash Equilibrium; A careful balancing point where each decision maker had made the best choice they could, given all other decision makers would do the same.

I never developed much intuition through the math we learned. But the question of incentives resonated. Even now when I observe a surprising behavior, I imagine myself in the other person’s shoes. I ask myself what perspective would make their choices sensible.

Objectives

I recently learned that George Dantzig, one of the most well-known names in operations research, felt his most valuable contribution to the field was the idea of the objective function. In optimization, you state the goals for the model, which is called the objective. There are other components of the model, but the objective is key.

Incentives in game theory are similar. Just like the best solution to an optimization problem will maximize the objective, a rational decision maker will maximize whatever they are incentivized to through their choices.

Metrics

Some of my motivation for today’s post came from reading this blog post on the idea of goals and the gaps that can occur between reality and a metric. Section 3 shares some novel and other well-known examples of the extremes that occur when you commit to maximizing some goal (for example, the AI that maximizes paper clip production by turning everything in the universe into paperclips).

Ultimately, if you define success, you should expect any decision-making entity to maximize the metric. Unfortunately, true success is typically a complex balancing act of many competing goals. Get the balance in your math slightly off from reality, and you should expect good metrics, and low success.

Minimizing the gap

I work with my clients regularly on defining success. One of my skills as a consultant is to predict how a particular metric will ultimately change the business compared to an alternative.

Say someone has declared maximizing profits to be their goal. Is that short-term profits, or over the entire future of the company? The best way to maximize really short-term profits might be to fire everyone and just sell what you have left on the shelf. The best way to maximize long-term profits will typically get the CEO fired since investing in the future takes time to prove it was done well.

It’s usually best to take a middle view. Understand the criteria investors / the board are using to measure success and do what we can to hit those metrics. For most organizations that strategy will give you a lower bound of short-term performance. You then want to optimize your ability through choices today to do the same for the months, quarters, and years ahead.

It is easy to be fooled as to how well you’re really doing if you don’t have good metrics. But consider the best choices individuals will make in pursuit of whatever goals they have been given. If your incentives don’t align to what you’re really trying to accomplish, you should expect bad outcomes.