Goodhart's Law: When a Metric Becomes a Target, It Stops Measuring Anything

Table of contents

• 1. From monetary policy to audit culture: the origins of Goodhart's law
• 2. Why metrics corrupt: the psychology of gaming
• 3. From banking to healthcare: 4 documented cases of perverse incentives
• 4. Comparison table: common ways performance metrics are gamed
• 5. Avoiding the trap: 4 safeguards to design reliable metrics
• 6. Protect your organization: a 5-step KPI audit
• 7. Conclusion: measuring to understand rather than to coerce
• 8. Staying on course with FluidOps
• Frequently asked questions

1. From monetary policy to audit culture: the origins of Goodhart's law

In July 1975, at the Conference in Monetary Economics hosted in Sydney by the Reserve Bank of Australia, a forty-year-old British economist took the floor. Charles Albert Eric Goodhart, then a monetary policy adviser at the Bank of England and later a professor at the London School of Economics, was dissecting a glaring failure of the British Treasury. In his paper titled Problems of Monetary Management: The U.K. Experience, he articulated a sobering observation that would fundamentally reshape organizational sociology: as soon as policymakers attempt to target a specific statistical aggregate (such as the broad money supply M3) to rein in inflation by decree, the historical relationship between that aggregate and price stability vanishes overnight (Goodhart, 1975).

Charles Goodhart's reasoning was purely systemic: commercial banks and financial market actors responded rationally to regulatory pressure. To circumvent credit limits and protect their margins, they engineered new financial instruments and parallel liquidity structures outside the scope of the targeted metric. The monitored M3 indicator displayed apparent restraint, yet underlying inflation continued to mount unabated. Goodhart formalized his insight: any observed statistical regularity will tend to collapse once pressure is placed upon it for control purposes (Goodhart, 1975).

This principle quickly expanded beyond central banking. In 1979, American psychologist and social scientist Donald T. Campbell independently formulated what research now terms Campbell's law: the more any quantitative social indicator is used for social decision-making, the more subject it will be to corruption pressures and the more apt it will be to distort the very social processes it was designed to monitor (Campbell, 1979). Both frameworks converge on a shared reality: human beings will always optimize their behavior to fit the rules of the system in which they operate.

The definitive formulation widely quoted by managers today emerged in 1997 through British anthropologist Marilyn Strathern. Analyzing the pervasive rise of evaluation culture in UK universities, she condensed Goodhart's insight into a memorable adage: "When a measure becomes a target, it ceases to be a good measure" (Strathern, 1997). When a thermometer is transformed into an instrument of punishment or reward, teams quickly learn to dip the thermometer into warm water instead of treating the fever.

2. Why metrics corrupt: the psychology of gaming

To grasp the pervasive impact of Goodhart's law across modern organizations, one must discard a persistent misconception: metric manipulation is rarely the result of malice or personal laziness. On the contrary, metric gaming is often the most rational response of competent professionals navigating poorly calibrated incentives. As early as 1956, management scholar Valentine F. Ridgway warned organizations about the dysfunctional consequences of raw individual performance measurement (Ridgway, 1956).

Confusing proxy metrics with operational reality

In any organization, what holds genuine value is almost always complex, qualitative, and hard to quantify: enduring customer satisfaction, architectural resilience in source code, or psychological safety within a team. Because leaders cannot directly measure these intangible qualities, they rely on quantified stand-ins known as proxies:

• Average handle time to gauge customer support efficiency.
• Closed tickets to assess technical team productivity.
• Outbound calls logged per day to evaluate sales tenacity.
• Lines of code written to quantify software development output.

As long as these proxies remain passive observational signals, the underlying correlation holds reasonably well. The breakdown happens the exact moment leadership converts that proxy into a performance target tied to carrots or sticks (Muller, 2018). From that point forward, employees stop focusing on the core objective—resolving customer issues or fixing system defects—and focus strictly on maximizing the numeric tally that drives their bonus, performance rating, or job security.

The sacrifice of unmeasured work

This dynamic finds formal grounding in the multi-task principal-agent model developed by Bengt Holmström and Paul Milgrom (1991). Workers typically balance several dimensions in their day-to-day work: operational speed on one side, and quality, knowledge sharing, collaboration, and risk mitigation on the other. If leadership evaluates and compensates individuals exclusively on speed, economic theory demonstrates that workers will reallocate their full attention to speed while actively sacrificing unmeasured quality (Holmström & Milgrom, 1991). What gets unmeasured simply gets abandoned.

3. From banking to healthcare: 4 documented cases of perverse incentives

Management by raw numbers does not just undermine small teams. Academic literature and legal records document staggering cases where an uncritical devotion to metrics compromised world-renowned institutions.

The Wells Fargo unauthorized accounts scandal (2002–2016)

The retail banking scandal at Wells Fargo showcases the organizational failure that ensues when a sales KPI turns into dogma. Executive leadership instituted an aggressive cross-selling initiative titled "Eight is Great": every retail banking customer was expected to hold an average of eight financial products (checking accounts, savings plans, credit cards, insurance packages, and lines of credit). Severe daily quotas were imposed on branch staff, backed by explicit threats of public shaming or termination.

Unable to generate that demand organically from ordinary consumers, employees adapted to the incentives: they opened more than 3.5 million deposit and credit card accounts without customer knowledge or consent. When the scheme unraveled, the institution dismissed over 5,300 branch employees and ultimately agreed to a 3 billion dollar settlement with the U.S. Department of Justice in 2020 (Department of Justice, 2020). The cross-selling ratio, designed to monitor customer loyalty, shattered the institution's century-old standing.

Ambulance queuing in the British NHS

During the early 2000s, the UK Department of Health attempted to curb emergency department overcrowding by introducing a strict mandate: 98% of patients had to be seen, admitted, or discharged within 4 hours of arrival. Non-compliant hospital trusts faced severe financial sanctions.

In their comprehensive study of the English public healthcare system, Gwyn Bevan and Christopher Hood documented striking operational distortions (Bevan & Hood, 2006). To prevent the four-hour clock from starting, hospital staff kept patients waiting outside inside parked ambulances (a practice known as ambulance queuing), as time spent in the vehicle did not count toward official emergency intake records. Other facilities quickly wheeled patients into hallways renamed "observation units" solely to stop the administrative timer, without speeding up clinical care by a single second.

Standardized testing manipulation in Chicago Public Schools

The education sector offers an equally telling case study. When Chicago public schools introduced punitive policies tied to student scores on standardized tests—including school closure threats and teacher salary freezes—educator behavior adapted accordingly. Economists Brian Jacob and Steven Levitt designed an innovative statistical algorithm to examine student-level answer sheets across tens of thousands of classrooms (Jacob & Levitt, 2003).

Their findings showed that active teacher cheating occurred annually in at least 4 to 5% of elementary classrooms (Jacob & Levitt, 2003). Certain educators erased incorrect pencil marks after testing sessions to insert correct responses, or fed answers directly to students while administering the exam. The summative assessment, meant to reflect academic mastery, ended up measuring teachers' fear of institutional retribution.

Risk avoidance in cardiac surgery report cards

In New York and Pennsylvania, health authorities mandated the public release of surgeon-specific report cards documenting post-operative mortality rates in coronary surgery, aiming to inform patients and encourage surgical excellence. Research conducted by David Dranove and colleagues revealed a tragic unintended outcome: surgeons intent on protecting their public ratings began systematically declining to operate on the highest-risk, most critically ill patients (Dranove et al., 2003).

Severely compromised patients were diverted toward conservative medical therapies that posed zero risk to a practitioner's public score, but were frequently less effective. The study demonstrated that introducing these public metrics paradoxically harmed overall clinical outcomes and drove an increase in mortality among the most vulnerable patient populations (Dranove et al., 2003).

4. Comparison table: common ways performance metrics are gamed

In his influential sociological work, Jerry Z. Muller highlights how the "tyranny of metrics" stems from an unquestioning faith in numbers as an automatic replacement for human judgment and field expertise (Muller, 2018). The table below outlines common gaming behaviors documented in economic and organizational research:

To learn more about structuring dashboards that reflect actual project progress without devolving into numerical micromanagement, see our guide on suivi de projet : rituels et indicateurs.

5. Avoiding the trap: 4 safeguards to design reliable metrics

Acknowledging Goodhart's law does not mean giving up on measurement altogether. Operating an organization or project without data is the equivalent of flying blind through a storm, leaving strategic decisions to personal bias or favoritism (Muller, 2018). Proven approaches exist to utilize performance metrics without triggering systematic distortion.

1. Always pair metrics with counterbalancing indicators

An isolated indicator invites unintended gaming. Whenever you establish a throughput or velocity metric, pair it with a rigorous qualitative safeguard. For instance, if you measure features delivered per sprint, pair that volume with production defect rates and customer Net Promoter Scores. Teams cannot artificially rush delivery if doing so immediately triggers alarms on the balancing metric.

2. Decouple diagnostic metrics from direct financial rewards and penalties

Donald Campbell emphasized that the vulnerability of any metric escalates with the stakes attached to it (Campbell, 1979). Operational diagnostics—such as analyzing pipeline bottlenecks or conversion drop-offs—must remain safe spaces for organizational learning. The moment a diagnostic metric dictates bonus payouts or job termination, it forfeits its observational integrity.

3. Track end-to-end system flow over individual output quotas

Instead of logging hours worked or counting tasks completed per individual, direct your attention toward collective workflow health. In agile environments, tracking overall lead time and establishing work-in-progress (WIP) limits prevents individualistic corner-cutting. You can draw on the méthode Kanban to visualize continuous team flow without singling out individual operators.

4. Preserve qualitative judgment and direct field observation

No dashboard or spreadsheet can replace first-hand observation. Japanese manufacturing philosophy formalized this principle as Gemba: going directly to where work happens to listen to front-line teams. When statistical variances emerge, investigate the operational context before assuming a deficiency in effort.

6. Protect your organization: a 5-step KPI audit

If you suspect that certain performance targets are driving counterproductive habits in your team, use this pragmatic five-step audit to recalibrate your metrics:

• Step 1: Map your active KPIs. Document all formal performance metrics monitored in your department or business, noting which indicators determine bonuses, rankings, or performance reviews.
• Step 2: Pose the benign gaming question. Bring your team together in a high-trust session and ask candidly: "If you needed to boost this metric by 40% next month without working additional hours or improving customer value, how would you do it?" The immediate responses will reveal the structural blind spots in your tracking system.
• Step 3: Uncover overlooked responsibilities. For each output metric, identify the critical unmeasured contributions (mentoring, cross-team support, documentation, preventive maintenance) that current dashboards ignore.
• Step 4: Apply the 80/20 rule. Eliminate metrics that exist merely for bureaucratic compliance without informing strategic choices. To separate meaningful data from background noise, reference la loi de Pareto en PME.
• Step 5: Replace numerical quotas with objective completion criteria. Rather than mandating call volumes, review the structured execution of client accounts. To prioritize work without succumbing to false urgency, examine la matrice d'Eisenhower au crible de la psychologie.

7. Conclusion: measuring to understand rather than to coerce

Goodhart's law is not an unavoidable operational fatality, but an essential call for managerial humility. Quantitative metrics make effective servants but destructive masters. They offer discrete signals meant to spark inquiry and open conversations, not definitive verdicts that shut down dialog.

By centering project management on human judgment, true delivery quality, and operational reality, you disarm gaming behaviors. Your performance measures can return to their legitimate purpose: serving as steady navigational compasses for your teams, rather than arbitrary constraints that restrict their potential.

8. Staying on course with FluidOps

To run your operations without falling into the trap of micromanagement and metric inflation, FluidOps centralizes your team's essential workflows in an intuitive, balanced workspace. Teams track real delivery milestones using the Kanban board, map critical project timelines with the diagramme de Gantt, and organize sprint cadences through the Scrum view, all while syncing deadlines directly with Google Calendar and Outlook. The Eisenhower matrix and other workflow optimization tools are available on our free tier to help your organization focus on high-impact work over vanity metrics.

FluidOps scales transparently with your organizational needs: the Free plan is available at €0 per month, the Individual plan is priced at €12 per month or €120 per year, and the Enterprise plan starts at €12 per seat per month on an annual billing cycle. Review all tier details on our page des tarifs.

Frequently asked questions

What is the precise definition of Goodhart's law?

Originally formulated by economist Charles Goodhart in 1975 and popularized by Marilyn Strathern, the principle states that when a statistical measure is chosen as a performance target, it ceases to be an accurate measure, because individuals optimize the metric itself rather than the underlying intent.

What is the difference between Goodhart's law and Campbell's law?

Goodhart's law originated in monetary policy and financial economics, whereas Donald Campbell's law (1979) emerged from social psychology and public policy evaluation. Both frameworks reach the same conclusion: using quantitative indicators to reward or punish actors inevitably corrupts the measure and distorts institutional processes.

How can you tell if a KPI has fallen victim to Goodhart's law?

The most common warning sign is a dashboard where every indicator shows green while operational outcomes deteriorate (such as unhappy clients, employee turnover, or rising production defects). This divergence indicates that metric optimization has disconnected from real-world performance.

Should organizations abandon all KPIs to prevent metric gaming?

No. Eliminating all metrics leads to arbitrary management decisions driven by unchecked bias or intuition. The appropriate solution is to deploy balanced pairs of countervailing metrics (quantity versus quality) and use KPIs primarily for operational diagnosis and continuous learning rather than punitive appraisals.

Sources

• Bevan, G., & Hood, C. (2006). What’s measured is what matters: Targets and gaming in the English public health care system. Public Administration, 84(3), 517–538. doi:10.1111/j.1467-9299.2006.00600.x
• Biagioli, M., Kenney, M., Martin, B. R., & Walsh, J. P. (2019). Academic misconduct, misrepresentation and gaming: A reassessment. Research Policy, 48(2), 401–413. doi:10.1016/j.respol.2018.10.025
• Campbell, D. T. (1979). Assessing the impact of planned social change. Evaluation and Program Planning, 2(1), 67–90. doi:10.1016/0149-7189(79)90048-X
• Courty, P., & Marschke, G. (2004). An empirical investigation of gaming responses to explicit performance incentives. Journal of Labor Economics, 22(1), 23–56.
• Department of Justice. (2020). Wells Fargo agrees to pay 3 billion dollars to resolve criminal and civil investigations into sales practices. U.S. Department of Justice Office of Public Affairs.
• Dranove, D., Kessler, D., McClellan, M., & Satterthwaite, M. (2003). Is more information better? The effects of "report cards" on health care markets. Journal of Political Economy, 111(3), 555–588. doi:10.1086/374180
• Eterno, J. A., & Silverman, E. B. (2012). The crime numbers game: Management by manipulation. CRC Press.
• Goodhart, C. A. E. (1975). Problems of monetary management: The U.K. experience. In Papers in Monetary Economics (Vol. 1). Reserve Bank of Australia.
• Holmström, B., & Milgrom, P. (1991). Multitask principal-agent analyses: Incentive contracts, asset ownership, and job design. Journal of Law, Economics, & Organization, 7, 24–52.
• Jacob, B. A., & Levitt, S. D. (2003). Rotten apples: An investigation of the prevalence and predictors of teacher cheating. The Quarterly Journal of Economics, 118(3), 843–877. doi:10.1162/00335530360698441
• Muller, J. Z. (2018). The tyranny of metrics. Princeton University Press.
• Ridgway, V. F. (1956). Dysfunctional consequences of performance measurements. Administrative Science Quarterly, 1(2), 240–247. doi:10.2307/2390989
• Strathern, M. (1997). 'Improving ratings': Audit in the British University system. European Review, 5(3), 305–321. doi:10.1002/(SICI)1234-981X(199707)5:3<305::AID-EURO184>3.0.CO;2-4