What AI Can Teach Us About Designing Better KPIs

The way organizations measure their performance is crucial to their success. The lessons we derive from technology and artificial intelligence offer valuable insights into improving these performances, aiming to avoid undesirable behaviors and achieve sustainable results.

Summary

  • Flawed performance measurement led to scandals such as those at Wells Fargo.
  • Goodhart’s law illustrates the risks of focusing on a single metric.
  • AI techniques provide innovative solutions for the design of KPIs and performance metrics.

The Lessons from Wells Fargo

In 2016, Wells Fargo faced severe criticism when it was revealed that employees, under pressure from aggressive sales targets, had opened billions of unauthorized customer accounts. This unfortunate situation was not only the result of unethical behavior but also of a flawed approach to performance measurement. Management encouraged employees to sell eight financial products per customer, stimulating undesirable and harmful behaviors.

The Dangers of Goodhart’s Law

Goodhart’s law states that when a metric becomes a target in itself, it loses its value as a measurement tool. Organizations worldwide suffer from a fixation on metrics, leading to the pursuit of narrow indicators that promote gaming and ethical violations.

AI Insights for Better KPI Designs

The insights from AI researchers can help organizations solve entrenched measurement problems. When training AI models, it is essential that they are not only focused on optimizing proxy measures but also understand the underlying goals.

A New Lens: Insights from AI Training

AI researchers study the phenomenon of overfitting, where models perform well on training data but fail on new data. These lessons can help organizations design KPIs more effectively.

The Metric Intelligence Framework in Action

Machine learning has developed four strategies to combat overfitting. These strategies have direct implications for how organizations can measure their performance.

1. Early Stopping

By continuously reassessing performance, organizations can avoid over-optimization. Amazon applies this principle by regularly reviewing its performance metrics.

2. Noise Injection

By introducing controlled randomness, organizations can build robustness. Random audits and rotating team members are effective ways to counter predictable behavior.

3. Capacity Alignment

Organizations must align the complexity of their metrics with their capabilities. A company with advanced analytical infrastructure can implement more complex, multidimensional metrics.

4. Regularization

This technique helps shift the focus from merely proxy measurements to broader goals. By imposing constraints, organizations can reduce the gaming behavior that arises from over-optimization.

The Future of Intelligent Measurement

By combining AI research with organizational design, companies can develop more effective measurement strategies. This requires continuous refinement of the measurement process and collaboration with employees who will use the systems.

Implementing these strategies can help organizations not only design better KPIs but also enhance recruitment and executive search practices. This enables them to attract the right people who align with their long-term goals.

Source: Balázs Kovács, January 21, 2026, https://sloanreview.mit.edu/article/what-ai-can-teach-us-about-designing-better-kpis/

Scroll to Top