Goodhart’s law says that once a measure is used as a target, it stops being a good measure, because people learn to move the number rather than the thing it was meant to track. Economist Charles Goodhart stated the original version in a 1975 paper on UK monetary policy: “any observed statistical regularity will tend to collapse once pressure is placed upon it for control purposes.” The popular wording, “when a measure becomes a target, it ceases to be a good measure”, is credited to anthropologist Marilyn Strathern (1997).
Any proxy carries noise, so picking the top scorers also picks the luckiest ones.
02
Extremal
Pushing a metric far enough moves you into territory where the old link between metric and goal no longer holds.
03
Causal
The act of intervening on the metric breaks the relationship that made it useful.
04
Adversarial
Someone with different goals learns the metric and games it. These four types come from Manheim and Garrabrant (2018).
Where you see it today
Reward hacking in AI
Researchers call it specification gaming: a system satisfies the literal objective without achieving the intended outcome. DeepMind researchers collected many examples from reinforcement learning (Krakovna et al., 2020).
Chat models are tuned against a reward model that predicts human preferences. Pushing that proxy too hard can make true quality worse, a Goodhart effect that researchers measured in a controlled setup (Gao et al., 2022).
Leaderboards as targets
Public AI leaderboards shape which models get attention. A 2025 study of Chatbot Arena argued that private testing of many variants and selective disclosure of the best scores bias the rankings (Singh et al., 2025).
Engagement metrics
A review of real-world cases found that optimising single metrics such as clicks or watch time leads to gaming and short-term focus, and proposed using a slate of metrics, qualitative accounts and the people affected (Thomas and Uminsky, 2020).
Goodhart's law vs related ideas
Goodhart's law
What it saysA measure used as a target stops being a good measure
Typical settingMonetary policy, KPIs, AI training
Campbell's law
What it saysA close cousin about social indicators used for decisions, arguably earlier
Typical settingPublic services and schools
Lucas critique
What it saysRelationships seen in past data shift once policy acts on them
Typical settingMacroeconomic policy
Cobra effect
What it saysAn incentive ends up making the problem worse
Typical settingBounties and subsidies
Specification gaming
What it saysAn AI system meets the literal objective but misses the intent
Typical settingReinforcement learning and agents
How to use it
For every target, write down the outcome it stands for and check that outcome directly from time to time.
Use a small set of metrics that pull against each other, such as speed and quality.
Pair numbers with qualitative review: read the tickets, calls or outputs behind the score.
When an AI model or vendor tops a benchmark, test it on your own tasks before trusting the rank.
Rotate or refresh metrics once teams have learned to optimise them.
The Business Engineer lens
AI systems are built by optimising numbers, so Goodhart’s law is now an engineering problem as much as a management one. The Business Engineer’s view is that whatever can be benchmarked tends to get automated, which makes the choice of benchmark a strategic decision about what gets built.
When a number becomes a target, people start optimising the number instead of the goal, so the number stops telling you what it used to.
Who came up with Goodhart's law?
Economist Charles Goodhart, in a 1975 paper on UK monetary policy; the popular phrasing is credited to anthropologist Marilyn Strathern (1997).
How does Goodhart's law apply to AI?
Models are trained against proxies such as reward models and benchmarks; optimising those proxies too hard can lead to reward hacking and inflated leaderboard scores.
How do you avoid Goodhart's law?
Use several metrics that balance each other, check the real outcome directly, add qualitative review, and refresh metrics once they are being gamed.
The complete explainer, cleaned up and reformatted for mobile.
Goodhartโs Law is named after British monetary policy theorist and economist Charles Goodhart. Speaking at a conference in Sydney in 1975, Goodhart said that โany observed statistical regularity will tend to collapse once pressure is placed upon it for control purposes.โ Goodhartโs Law states that when a measure becomes a target, it ceases to be a good measure.
Concept Overview
Explanation– Goodhart’s Law is a fundamental concept in economics and social sciences that highlights a potential pitfall when using metrics or indicators as a basis for decision-making. The law states: “When a measure becomes a target, it ceases to be a good measure.” In other words, when people or organizations set specific metrics or indicators as targets to achieve, they may change their behavior in ways that undermine the original purpose of the measure. Goodhart’s Law was formulated by economist Charles Goodhart and has implications in various fields, including finance, economics, business, and public policy.
Key Principle
Explanation– The key principle of Goodhart’s Law is that using a metric as a target can lead to unintended consequences. When individuals or organizations are incentivized to achieve a particular metric, they may focus solely on optimizing that metric, even if it comes at the expense of other important factors or the original goal. This can result in distorted behavior, misaligned incentives, and outcomes that are counterproductive or contrary to the intended purpose.
Examples
Explanation– Examples of Goodhart’s Law in action include: 1. Financial Markets: When stock prices or earnings per share (EPS) become the primary focus for a company, management may engage in financial engineering or short-term strategies to boost these metrics, potentially at the expense of long-term growth and sustainability. 2. Education: In education systems where standardized test scores are used as a primary metric for evaluating schools or teachers, educators may “teach to the test” rather than providing a well-rounded education. 3. Healthcare: In healthcare, if hospital performance is measured based on patient satisfaction scores, healthcare providers may prioritize patient comfort over necessary medical procedures or treatments. 4. Corporate Performance Metrics: When organizations set specific quarterly revenue targets as key performance indicators (KPIs), sales teams might push for aggressive sales tactics, discounting, or revenue recognition manipulation to meet the targets.
Applications
Explanation– Understanding Goodhart’s Law has applications in various fields: 1. Business Metrics: Business leaders should be cautious about relying solely on a single metric as a performance indicator and consider potential unintended consequences. 2. Public Policy: Policymakers need to design policies and regulations that account for the possibility of metrics being manipulated or distorted. 3. Financial Markets: Investors and analysts should consider how companies may be influenced by financial metrics in their investment decisions. 4. Education and Healthcare: Administrators and policymakers should use a balanced set of metrics to evaluate performance rather than relying solely on one measure.
Mitigation Strategies
Explanation– To mitigate the negative effects of Goodhart’s Law, individuals and organizations can consider the following strategies: 1. Diversify Metrics: Use a combination of metrics rather than relying on a single measure to assess performance. 2. Scrutinize Incentives: Carefully examine the incentives created by performance metrics to ensure they align with the intended goals. 3. Regularly Review Metrics: Continuously assess whether the chosen metrics are still relevant and meaningful for evaluating performance. 4. Encourage Transparency: Foster a culture of transparency and open communication to reduce the likelihood of gaming the metrics. 5. Monitor for Unintended Consequences: Actively monitor for any unintended consequences or distortions resulting from the use of metrics as targets.
Impact and Relevance
Explanation– Goodhart’s Law is significant because it underscores the complexity of human behavior and decision-making in various contexts. It serves as a reminder that setting targets based on metrics requires careful consideration and monitoring to avoid perverse incentives and unintended outcomes. The concept has broad implications in finance, economics, education, healthcare, and public policy, and it continues to be a critical element in decision-making and performance evaluation worldwide.
Understanding Goodhartโs Law
Goodhart would later admit that his quip was intended to be a humorous, throw-away comment. But it was nevertheless an accurate and perceptive observation about how the modern world functions.
Itโs important to note that Goodhart himself had no role in naming the law for which he is named. That distinction goes to anthropologist Marilyn Strathern, who argued in a 1997 paper that the law had uses beyond statistics to evaluation in a broader sense.
An oft-told story of Goodhart’s Law at work can be described by the cobra effect. In India under British colonial rule, the government was troubled by the number of venomous cobras. To reduce their population, the government placed a bounty on every cobra the locals could catch. This strategy worked for a while, but some individuals began breeding the cobras only to kill them later and collect a higher bounty.
Eventually, the colonial government caught on and scrapped the scheme, causing many of the bred cobras to be released into the wild. The key takeaway of the cobra effect story is that incentives designed to solve a problem end up rewarding people for making the problem worse.
The four forms of Goodhartโs Law
There are generally accepted to be four variations on Goodhartโs Law
Regressive Goodhart โ here, the measures individuals use for their target (goal) are imperfectly correlated with that goal. For example, weight is imperfectly correlated with health because it encourages skipping meals or weighing oneself in the morning with an empty stomach.
Extremal Goodhart โ this occurs when a measurement is picked because it correlates with a goal in normal situations. In extreme circumstances however, the measure is erroneous. The human relationship with sugar is a classic example. While sugar was correlated with survival thousands of years ago, the same cannot be said of modern, sedentary lifestyles where sugar promotes obesity.
Causal Goodhart โ where the behavior of an individual does not directly affect the goal but has some causal effect on the measure. The number of times a gym membership is renewed does not directly impact how often an individual exercises, for example.
Adversarial Goodhart โ where other goals confound the goal a measure is trying to accomplish, such as the cobra effect mentioned above.
Avoiding the impact of Goodhartโs Law
Of the four variations of Goodhartโs Law, only the Regressive Goodhart is unavoidable.
For the remaining three, here are some simple avoidance tips
Conduct regular checks to ensure the measure is still incentivizing in line with the desired outcome or goal.
Become aware of Goodhartโs Law and how it operates.
Maintain a focus on the end goal while using the measures as a guide only.
Reduce bureaucracy and formalism.
Use a combination of diversified metrics. A balanced scorecard can be useful here.
When and How Goodhart’s Law Manifests
Goodhart's Law can manifest in various scenarios
Economic Policies: When governments set specific economic targets, individuals and businesses may adjust their behavior to meet those targets, potentially distorting the economy.
Financial Markets: In financial markets, traders and investors may manipulate stock prices or financial ratios to achieve short-term gains.
Education: Educational institutions may focus excessively on standardized test scores, leading to teaching strategies geared toward test preparation rather than holistic learning.
Business Metrics: Companies can fall into the trap of optimizing key performance indicators (KPIs) at the expense of long-term sustainability or customer satisfaction.
To understand and address Goodhart's Law
Diverse Metrics: Utilize a diverse set of metrics and indicators to assess performance, rather than relying solely on one measure.
Monitoring and Adaptation: Continuously monitor for unintended consequences and adapt metrics and targets accordingly.
Contextual Understanding: Consider the broader context and incentives surrounding the use of metrics to anticipate potential distortions.
Periodic Reevaluation: Regularly review and update the metrics used for assessing performance to account for changing circumstances.
Implications and Drawbacks of Goodhart’s Law
Goodhart's Law carries several implications and potential drawbacks
Goal Distortion: The primary drawback is the distortion of goals, where individuals or entities focus on optimizing the metric itself rather than the underlying objective.
Unintended Consequences: Overemphasis on a single metric can lead to unintended negative consequences, such as neglecting other critical factors or engaging in unethical behavior.
Short-Term Focus: Entities may prioritize short-term gains to meet immediate targets, potentially sacrificing long-term sustainability.
Measurement Complexity: Managing multiple metrics can be more complex and resource-intensive than relying on a single indicator.
Difficulty in Identifying Metrics: Identifying suitable metrics that accurately capture desired outcomes can be challenging.
Examples and Case Studies
Academic Performance: In education, the use of standardized test scores as the primary measure of academic performance can lead to unintended consequences. Schools and teachers may be incentivized to focus solely on improving test scores, leading to teaching to the test and neglecting other important aspects of learning and development.
Business Metrics: In business, focusing solely on financial metrics like revenue or profit can lead to adverse effects. For example, if a sales team is rewarded solely based on revenue, they may engage in aggressive sales tactics or offer steep discounts to close deals, even if it negatively impacts long-term profitability.
Healthcare: In healthcare, the use of certain performance metrics to assess the quality of care can create unintended outcomes. For instance, a hospital that prioritizes reducing patient wait times may discharge patients prematurely or prioritize less critical cases to improve wait time metrics, potentially compromising patient outcomes.
Employee Performance: In the workplace, setting individual performance targets without considering the broader context can lead to unintended consequences. Employees may focus solely on meeting their targets, neglecting collaboration and teamwork, or engaging in unethical behavior to achieve their goals.
Government Policies: When governments set specific targets to measure the success of policies, such as reducing unemployment rates or increasing GDP growth, there is a risk of distorting efforts and focusing on short-term gains rather than addressing long-term challenges.
Key takeaways
Goodhartโs Law states that when a measure becomes a target, it ceases to be a good measure.
Goodhartโs Law was informally coined during a speech by Charles Goodhart. Although the economist was speaking in the context of statistics, the law has broader evaluative applications.
Goodhartโs Law is generally categorized into four variations: Regressive Goodhart, Extremal Goodhart, Causal Goodhart, and Adversarial Goodhart.
Key Highlights
Goodhart’s Law: Named after economist Charles Goodhart, the law states that when a measure becomes a target, it loses its effectiveness as a measure.
Understanding the Law: Despite Goodhart’s original comment being meant humorously, it accurately describes how measures can be manipulated when used as targets.
Origin of the Term: Anthropologist Marilyn Strathern is credited with naming the law and extending its relevance beyond statistics to broader evaluation contexts.
The Cobra Effect: An example illustrating Goodhart’s Law is the cobra effect, where government incentives to reduce cobra populations backfired when people started breeding them for the rewards.
Four Variations of the Law:
Regressive Goodhart: Measures used for goals have imperfect correlation with those goals.
Extremal Goodhart: Measures that correlate with goals in normal situations fail under extreme circumstances.
Causal Goodhart: An individual’s behavior indirectly affects the measure but not the actual goal.
Adversarial Goodhart: Conflicting goals interfere with the desired goal of the measure.
Avoiding Goodhart’s Law:
Regularly reassess measures to ensure they align with the intended outcomes.
Understand and acknowledge the operation of Goodhart’s Law.
Keep the focus on the end goal while using measures as guides.
Minimize bureaucracy and formality.
Use a combination of diverse metrics, such as a balanced scorecard.
Examples and Case Studies:
In education, focusing solely on standardized test scores can lead to neglect of broader learning aspects.
In business, prioritizing financial metrics can lead to short-term gains at the expense of long-term profitability.
In healthcare, emphasizing certain metrics can compromise patient care quality.
In the workplace, setting individual performance targets can undermine collaboration and ethical behavior.
In government policies, setting specific targets may lead to neglect of long-term challenges.
Key Takeaways:
Goodhart’s Law highlights the problem of using a measure as a target.
Named after Charles Goodhart, the law has broader applications beyond statistics.
The four variations of the law explain different ways the law manifests in various contexts.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
