In 2016, Wells Fargo faced a scandal where employees opened millions of unauthorized accounts due to pressure from aggressive sales targets. This highlighted how metrics can overshadow their true purpose, leading to unethical behavior. This issue is an example of Goodhart’s law: when a measure becomes a target, it stops being a good measure. Despite warnings, many organizations still fall into the trap of focusing on narrow indicators that lead to gaming and poor outcomes.
Traditional tools like KPIs often fail because they don’t fully address this challenge. Insights from AI research—particularly the concept of overfitting—offer fresh ways to design performance measures. Overfitting occurs when AI models perform well on training data but fail to generalize in real-world scenarios because they latch onto irrelevant patterns. Organizations face analogous problems when teams optimize for metrics rather than true goals.
AI addresses overfitting through four key strategies: early stopping (halting optimization at the right time), noise injection (adding randomness to avoid gaming), capacity alignment (matching model complexity to oversight capabilities), and regularization (simplifying incentives to prevent extreme optimization). When applied to business, these concepts translate into practices such as regular reassessment of metrics, random audits, aligning metrics with company capabilities, and balancing incentives to prevent distortion.
Examples include Amazon’s periodic metric reviews, JPMorgan Chase’s random compliance audits, Microsoft’s adaptive goal-setting, and Netflix’s multi-metric approach to content success. Companies like Intel and Unilever show how shifting focus from single metrics to balanced ones leads to better alignment with customer value and sustainability.
To avoid metric fixation, organizations should ask whether teams are gaming metrics, if outcomes beyond dashboards are suffering, and whether metrics communicate true purpose. By incorporating these AI-inspired practices, organizations can create more robust, meaningful performance measures that drive genuine progress while reducing harmful behaviors. This approach requires ongoing refinement, engagement with teams, and cultural shifts from short-term results to long-term value.
Ultimately, while AI principles enhance measurement, human judgment remains essential to balance structure with flexibility. In a world with increasing data and AI integration, these lessons are vital to designing KPIs that genuinely reflect and support organizational goals.
