6 Principles of Incentive Design
There are few forces in the world as powerful and ubiquitous as inspiration. They govern everything from our day-to-day interactions and decision-making to our wider organizations and societies.
Charlie Mungerzhong said it well:
"Give me motivation, and I will give you the effect."
Incentives seem to have a unique duality:
- Well-designed incentives can have wonderful results.
- Poorly designed incentives can have dire results.
Surprisingly, despite their importance, incentives are rarely studied in schools, business programs, or organizations.
So humans are still surprisingly bad at creating proper incentives. We have been creating systems that invite manipulation and open the door to unintended consequences. Too often, we fall into poorly designed camps and find ourselves scrambling for answers and quick fixes.
In today's article, I'll cover some common pitfalls in incentive design and propose a core set of principles for building thoughtful incentives that produce desired outcomes.
Let's start with a basic definition of motivation: An incentive is anything that motivates, inspires, or drives an individual to behave in a particular way.
They come in two forms:
- Intrinsic: Intrinsic - created by self-interest or desire.
- Extrinsic: Extrinsic - caused by external factors, usually rewards (positive incentives) or punishments (negative incentives).
Much of my writing here and on Twitter is loosely related to intrinsic motivation (our motivations, desires, and pursuits).
But today, I'm going to focus specifically on extrinsic motivation.
In a simplified general model, extrinsic motivation involves two key components:
Metric: A standard by which an individual or group is measured. Metrics can be quantitative (KPIs, indicators) or qualitative.
Target: The level of action that will initiate a reward or penalty. Goals can be specific (you will be rewarded if the KPI reaches X level) or general (you will be rewarded if your manager is happy with your work).
But there's a problem: This simple incentive model -- which should feel familiar -- often leads to bad outcomes and unintended consequences.
The "culprit" of today's article: Goodhart's Law.
Goodhart's law is simple:
When a measure becomes a goal, it ceases to be a good measure. If performance measurement becomes a given goal, humans tend to optimize for it without regard to any associated consequences.
Goodhart's Law is named after the British economist Charles Goodhart, who referenced the concept in a 1975 article on UK monetary policy.
"As soon as it is stressed for control purposes, any observed statistical regularity tends to collapse."
But the concept was popularized by anthropologist Marilyn Strathern. In a 1997 paper, she generalized this idea and called it Goodhart's Law.
"When a measure becomes a target, it ceases to be a good measure."
It quickly became a mental model of considerable practical interest—a phenomenon that describes (and accurately predicts) the failure of overly simplistic incentive schemes.
Let’s look at a few examples and use them to create a framework for where incentives go wrong.
Indian Cobra and Soviet Nails
India has too many cobras.
The British see cobra heads as a cleansing measure to get rid of cobras, so it encourages people to offer cobra heads.
result? Locals game the system, breeding cobras for bounty. Incentives designed to reduce cobra numbers actually increased it.
The USSR needed to produce more nails to fuel its military-industrial complex.
First, the Soviet leadership instituted incentives based on the number of nails produced. result? These factories produced thousands of tiny nails.
Next, leadership adjusted the incentives based on the weight of the nails produced. result? The factory produced some big nails.
In both cases, nails are useless.
Amazon believes employee turnover is a driver of long-term business success.
From the early days, Jeff Bezos created a culture that would eliminate the bottom 10% of employees in order to continuously improve the talent level of the organization.
To drive healthy employee turnover at scale, it sets annual turnover targets for managers.
result? Countless articles have appeared on the practice of "hire-to-fire". Managers allegedly hired employees they planned to fire to meet their turnover targets.
Wells Fargo Account Opening
Wells Fargo's senior leadership sees new account openings as an easy way to track business growth.
To drive new account openings, it set account opening goals for its junior employees. Employees will be forced to exceed these targets to earn their bonuses (or possibly be penalized if they miss).
result? Employees opened millions of fake accounts to hit targets, and Wells Fargo was fined billions of dollars for apparent fraud.
Framing of incentives
With these examples in the background, we can begin to develop a simple framework for understanding where incentives go wrong.
Poorly designed incentives typically exhibit one or more of the following three characteristics:
- McNamara Fallacy
- Vanity > Quality
The McNamara Fallacy is named after Robert McNamara, US Secretary of Defense from 1961 to 1968, whose over-reliance on quantitative metrics is widely believed to have led the US astray during the Vietnam War.
It is the tendency to make decisions based on observable quantitative metrics and ignore all others. It lets us focus on measuring what is easy to measure and what is really important (i.e. something that expresses a desired outcome).
Cobra heads, nail count or weight, employee turnover and new account openings are all easy to quantify, but completely ignore the bigger picture. All four incentive programs are victims of the McNamara fallacy.
The narrow focus problem is an objective scoping problem.
If you think too narrowly about the desired outcome of your program, you are more likely to create incentives that miss the forest for the trees.
In the case of Wells Fargo, the desired outcome is not opening more accounts at the bank—it is more appropriate to define the desired outcome as an increase in the number of satisfied, well-served customers.
Rule of thumb: when in doubt, zoom out.
Vanity > Quality
Relying on vanity metrics—such as cobra heads or new account openings—to impress superiors or the public is a recipe for disaster in incentive design.
Imagine incentivizing social media managers based on the number of followers an account has. That person may start buying followers to achieve these goals.
Vanity metrics are rarely quality metrics.
Incentive Design Principles
Building on our Fragmented Incentive Framework, let's establish thoughtful incentive design principles.
Six principles to consider when developing thoughtful incentives:
- inverse measure
- Stakes and Effects
- participate in the game
- clear and smooth
Let's look at each principle.
It is critical to think deeply about the ultimate goal of the incentive.
What does success look like? What is the final desired outcome?
It's not about superficial goals - you need to go deeper. Smart incentive design is impossible without deep thinking about goals.
Be sure to start here before moving on.
Establish metrics that you will measure to track success.
It’s important to avoid the McNamara fallacy — never choose a metric based on what’s easy to measure rather than what’s actually meaningful.
Just because it's easy to track a particular KPI doesn't mean it's the right KPI to use as a measure.
Ask yourself: If you could track and measure one metric that told you everything you ever wanted to know about your business or organization, what would it be?
A wishlist for determining metrics without regard to feasibility. From here, move on to what is possible.
Perhaps even more important than core metrics, build "counter metrics" of what you measure to track unintended consequences.
I was first introduced to this idea by Julie Zhou (whom she calls anti-metrics), who has done some extraordinary thinking and writing on the topic of organization and growth.
Counter metrics force you to consider whether your incentives solve one problem but create another.
In the case of Amazon, an effective counter-standard might be the average tenure of new hires by cohort. You'll know something is wrong if you see this number drop sharply from the start of the employee separation incentive plan.
Counter indicators will tell you if you've won the battle but lost the war.
Stakes and Effects
As with all decisions, it is critical to consider and understand the stakes:
- High risk = costly failure, hard to reverse
- Low risk = cheap failure, easy to reverse
If you are dealing with a high-risk procedure, you must perform a rigorous analysis of second-order effects.
Iterate over your indicators and counter indicators accordingly.
participate in the game
To avoid the principal-agent problem, incentive designers should be involved.
Incentives should never be allowed where creators participate in the fun rather than the pain.
Participation in the game improves outcomes.
clear and smooth
Incentives are only valid if:
- Its launch clarity.
- The ability and willingness to adjust to new information.
Takeaway: Create a more understandable playing field for all members and avoid plan continuation bias.
Incentives are the most powerful tool in a modern leader’s toolkit when designed properly.
To avoid incentive failure, please note the following:
- The McNamara Fallacy: The false assumption that something that cannot be measured is not important. Leads to a focus on measuring what is easy to measure and what really matters.
- Narrow focus: Thinking too narrowly about the intended outcomes of the program creates incentives that miss the forest for the trees.
- Vanity > Quality: Rely on vanity metrics that impress superiors or the public.
To create thoughtful incentives, focus on six principles of incentive design:
- Goal: Identify what success looks like. Learn more about the program's ultimate goals.
- Metrics: Establish metrics to track success. Never settle for something that is easy to measure and not something that really counts.
- Counter-indicators: Establish counter-indicators to determine whether solving one problem creates another.
- Risk and Impact: Always consider risk (high or low) and adjust the rigor of the second order impact analysis accordingly.
- Participation in the game: Avoid the principal-agent problem by ensuring that designers are incentivized to participate in the game (i.e., participate in both pleasure and pain).
- Clarity and Fluency: Incentives are only as good as the clarity with which they are rolled out and the ability to adjust to new information.
I hope you find this model as productive and useful as I did.