Incentive design as system change
~5 min read
Core Primitive
What gets measured and rewarded determines what people actually do.
Incentive design is the most powerful lever for systemic change because incentives operate continuously, automatically, and at scale — shaping behavior across the entire organization without requiring individual intervention. But incentives are also the most dangerous lever because poorly designed incentives produce precisely the behavior they measure, including the dysfunctional side effects of optimizing for the measured dimension at the expense of unmeasured dimensions. Goodhart's Law — "When a measure becomes a target, it ceases to be a good measure" — is the central challenge of incentive design.
The incentive machine
Charlie Munger, Warren Buffett's longtime partner, called incentives "the most powerful force in the world." He was not exaggerating. Incentives operate continuously — not just during performance review season but every hour of every day, shaping thousands of micro-decisions across the organization. They operate automatically — people do not need to consciously think about incentives for incentives to shape their behavior. And they operate at scale — a well-designed incentive system aligns the behavior of thousands of people without requiring individual management of each one.
The power of incentives is also their danger. A well-aligned incentive system produces remarkable organizational performance — behavior that is naturally directed toward organizational goals without requiring constant supervision. A misaligned incentive system produces equally remarkable dysfunction — behavior that is naturally directed away from organizational goals despite constant supervision.
Steven Kerr's seminal paper documented the ubiquity of incentive misalignment: organizations routinely reward behavior that contradicts their stated goals. They hope for long-term investment but reward quarterly results. They hope for teamwork but reward individual achievement. They hope for innovation but punish failure. They hope for quality but reward speed (Kerr, 1975).
Goodhart's Law and its consequences
Charles Goodhart, a British economist, observed that "when a measure becomes a target, it ceases to be a good measure." The observation, formalized as Goodhart's Law, captures the fundamental challenge of incentive design: the act of incentivizing a metric changes the behavior the metric was designed to measure.
Before a metric becomes a target, it accurately reflects the underlying reality it measures. Customer satisfaction scores reflect actual satisfaction. Code velocity reflects actual development speed. Employee engagement scores reflect actual engagement. But when the metric becomes a target — when people are rewarded for improving the number — the number becomes disconnected from the reality it was supposed to reflect.
People optimize for the metric, not for the underlying reality. Customer satisfaction scores improve because agents are coached to ask satisfied customers to rate their experience while avoiding the question with dissatisfied customers. Code velocity increases because engineers inflate story point estimates and split work into smaller tickets. Engagement scores improve because managers pressure team members to give positive responses. The metrics all look better. The underlying realities have not changed — or have gotten worse.
Marilyn Strathern generalized Goodhart's observation: "When a measure becomes a target, it ceases to be a good measure." The generalization applies beyond economics to any domain where targets are set and measured — education (teaching to the test), healthcare (optimizing measured outcomes while neglecting unmeasured patient needs), and organizational management (optimizing KPIs while neglecting the organizational health that the KPIs were supposed to indicate) (Strathern, 1997).
Principles of effective incentive design
Effective incentive design acknowledges Goodhart's Law and designs around it — creating incentive systems that are robust to gaming, aligned with organizational need, and adaptive to changing conditions.
Measure outcomes, not outputs
Outputs are the activities people perform. Outcomes are the results those activities produce. Measuring outputs (calls handled, tickets closed, features shipped) rewards activity regardless of impact. Measuring outcomes (problems resolved, customer retention, user adoption) rewards impact regardless of method.
The distinction matters because output metrics can be gamed without producing value (close more tickets by splitting each ticket into multiple sub-tickets), while outcome metrics can only be improved by actually producing value (retain more customers by actually solving their problems).
Use balanced scorecards, not single metrics
No single metric captures the full range of desired behavior. Every single metric creates pressure to optimize the measured dimension at the expense of unmeasured dimensions. A balanced scorecard — a set of three to five metrics that together capture the full range of desired behavior — prevents the lopsided optimization that single metrics produce.
Robert Kaplan and David Norton's balanced scorecard framework identifies four perspectives that together provide a comprehensive view: financial (are we creating value?), customer (are we serving our customers?), process (are our systems effective?), and learning (are we getting better?). A metric in each perspective prevents optimization in one perspective at the expense of others (Kaplan & Norton, 1992).
Include lagging and leading indicators
Lagging indicators (revenue, retention, satisfaction) measure outcomes that have already occurred. Leading indicators (pipeline quality, employee engagement, process health) predict outcomes that will occur. Incentivizing only lagging indicators rewards past behavior without influencing future behavior. Including leading indicators directs attention toward the activities that will produce future outcomes.
Design for intrinsic motivation
The strongest and most sustainable incentives are intrinsic — the satisfaction of doing meaningful work well. Extrinsic incentives (bonuses, promotions, recognition) are effective supplements to intrinsic motivation but poor substitutes for it.
Daniel Pink's research synthesized decades of motivation science into three conditions for intrinsic motivation: autonomy (the freedom to choose how to do the work), mastery (the opportunity to develop expertise), and purpose (the connection between the work and a meaningful goal). Incentive systems that support these three conditions amplify intrinsic motivation. Incentive systems that undermine them — by removing autonomy, preventing mastery, or disconnecting work from purpose — replace intrinsic motivation with extrinsic compliance, which is less durable and less energizing (Pink, 2009).
Build in anti-gaming mechanisms
Expect gaming and design for it. Common anti-gaming mechanisms include:
Peer assessment. Include peer evaluation alongside manager evaluation — peers often see gaming that managers miss because they observe behavior at a different angle.
Outcome auditing. Periodically audit the relationship between the metric and the underlying reality. If the metric improves but the reality does not, the metric is being gamed and should be redesigned.
Metric rotation. Periodically change which metrics are incentivized — not to keep people off-balance but to prevent the accumulation of gaming strategies that develop around stable metrics.
Qualitative overlay. Combine quantitative metrics with qualitative assessment — narrative performance evaluation that can capture dimensions the metrics miss.
The Third Brain
Your AI system can help you design and audit incentive systems. Describe the behaviors you want to encourage and the current metric and incentive structure, and ask: "Audit this incentive system for misalignment: Where does optimizing for the current metrics produce behavior that contradicts organizational goals? Where are unmeasured dimensions being neglected? Design an improved incentive system with a balanced scorecard (3-5 metrics across outcome, process, and health dimensions), leading and lagging indicators, and anti-gaming mechanisms."
From incentives to information
Incentives determine what people are motivated to do. Information determines what they are able to do. The next lesson, Information flow design, examines information flow design — how changing who gets what information and when changes organizational behavior as powerfully as changing what is measured and rewarded.
Sources:
- Kerr, S. (1975). "On the Folly of Rewarding A, While Hoping for B." Academy of Management Journal, 18(4), 769-783.
- Strathern, M. (1997). "'Improving Ratings': Audit in the British University System." European Review, 5(3), 305-321.
- Kaplan, R. S., & Norton, D. P. (1992). "The Balanced Scorecard — Measures That Drive Performance." Harvard Business Review, 70(1), 71-79.
- Pink, D. H. (2009). Drive: The Surprising Truth About What Motivates Us. Riverhead Books.
Put it into practice
Exercise
Audit the incentive system for one role in your organization. List every metric that is measured, reported, or rewarded — both formally (performance reviews, bonuses, promotions) and informally (what gets praised in meetings, what gets attention from leadership, what gets criticized). For each metric, ask: If someone optimized exclusively for this metric, what behavior would they exhibit? Would that behavior serve the organization well? If optimizing for the metric produces dysfunctional behavior, the metric is misaligned — it measures something different from what the organization actually needs. Redesign the metric to align measurement with organizational need. The most common misalignment: measuring output (how much was produced) when the organization needs outcome (what impact was achieved).
More on putting it into practice
See it in action
A call center, Respond, measured and rewarded agents on average handle time — the average duration of each customer call. The incentive produced exactly what it measured: agents rushed through calls, resolving the immediate question without addressing underlying issues, transferring complex cases to other departments rather than handling them, and sometimes disconnecting calls that would take too long. Average handle time decreased — the incentive was working. But customer satisfaction plummeted, repeat call rates increased (customers called back because their underlying issue was not resolved), and the transfer rate to other departments created bottlenecks that degraded service across the entire organization. The metric had become the target, and the target had ceased to be a good measure of what the organization actually wanted: customer problems resolved. Respond redesigned the incentive system: the primary metric became first-call resolution (was the customer's problem fully resolved in a single interaction?), with average handle time tracked but not incentivized. Agents began spending more time on calls — but repeat calls dropped 40%, transfers dropped 60%, and customer satisfaction scores increased by 25 points. The incentive redesign changed agent behavior not through training or motivation but through structural alignment: agents were now rewarded for doing what the organization actually needed.
Watch for the failure mode
Designing incentives that optimize one dimension at the expense of others. Every metric creates pressure toward the measured dimension and neglect of unmeasured dimensions. A sales team incentivized on revenue will pursue revenue at the expense of profitability. An engineering team incentivized on velocity will pursue speed at the expense of quality. A support team incentivized on response time will pursue speed at the expense of resolution. The failure mode is not that the incentive is wrong — it is that the incentive is incomplete. Single-metric incentives always produce lopsided optimization. Balanced incentives require multiple metrics that together capture the full range of desired behavior.
Make it stick
This lesson examines the most powerful structural lever introduced in Structural change versus behavioral change. Incentives are structures — they are part of the system's architecture, not separate from it. The connection to culture is built by repeated behavior is causal: incentives shape which behaviors are repeated, and repeated behaviors build culture. Change the incentives and you change the behavioral deposits, which changes the culture. The connection to What leaders tolerate defines culture more than what they praise (what leaders tolerate defines culture) is complementary: incentives define what is rewarded; tolerance defines what is permitted. Together, they set the behavioral range. Look forward to Information flow design, which examines information flow design — how changing who gets what information and when changes organizational behavior.
Practice
Map Dysfunctional Incentives for a Key Role in Miro
Audit and visualize the complete incentive system for one organizational role, identifying where measured metrics drive dysfunctional behavior and redesigning those metrics to align with actual organizational needs.
- 1Open Miro and create a new board titled 'Incentive System Audit: [Role Name]'. Create three swim lanes labeled 'Formal Incentives' (performance reviews, bonuses, promotions), 'Informal Incentives' (what gets praised/criticized), and 'Hidden Incentives' (what gets leadership attention). In each lane, add sticky notes for every metric that drives behavior for this role.
- 2For each metric sticky note in Miro, create a connected sticky note asking 'If optimized exclusively, what behavior results?' Use Miro's connector tool to link each metric to its predicted behavior. Be brutally specific: if the metric is 'tickets closed per week,' the behavior might be 'close easy tickets first, avoid complex problems, mark tickets resolved prematurely.'
- 3Use Miro's color coding to mark each metric: green for aligned (optimization serves the organization), red for misaligned (optimization produces dysfunction), yellow for unclear. Add a comment to each red sticky explaining the specific dysfunction: what unmeasured dimension gets sacrificed when this metric is optimized (e.g., 'speed' measured while 'quality' unmeasured).
- 4For each red (misaligned) metric in Miro, create a new sticky note in a fourth swim lane labeled 'Redesigned Metrics.' Transform output metrics into outcome metrics: instead of 'tickets closed,' try 'customer problems resolved (verified by follow-up)' or 'reduction in repeat tickets.' Use Miro's voting feature to mark which redesigned metrics would most improve the system.
- 5Create a Miro text box summarizing the three most dangerous misalignments discovered and their redesigned alternatives. Export this summary as a PDF and identify one redesigned metric you can pilot with stakeholders in the next two weeks, noting specifically what behavior change you expect to observe when measurement shifts from output to outcome.