All notes
1 min read

The Metric That Ruins Everything

How Goodhart's Law Quietly Ruins Every Product It Touches

TL;DR

When a proxy measure becomes the goal, the actual goal collapses, turning time-on-page into doomscrolling and engagement into outrage.

Software teams track everything now. Time spent on page. Click-through rates. Conversion funnels. Engagement scores. Heat maps showing where eyes linger. Every interaction becomes data, every behaviour a metric to optimize.

Then someone decides which metric matters most. That’s when things break.

Consider a news app optimizing for “time spent reading.” Sounds reasonable. More time reading means more engaged readers, right? So designers add infinite scrolling. Autoplay videos. Related article suggestions that never end. Clickbait headlines engineered to keep you scrolling.

The metric improves. Time spent skyrockets. But nobody’s actually reading anymore. They’re scrolling, skimming, chasing the next headline. The proxy measure (time spent) became disconnected from the actual goal (informed readers).

This pattern repeats across products. A shopping app focuses on conversion rates, so it adds aggressive pop-ups and manipulative countdown timers. Conversions increase. Customer satisfaction craters. The metric wins. The experience loses.

Social platforms optimize for engagement, which sounds innocuous until you realize engagement means any interaction. Rage-clicking counts. Hate-sharing counts. Doomscrolling at 2 AM counts. The algorithm doesn’t distinguish between “engaged because delighted” and “engaged because enraged.” It just wants you clicking.

Request an AI summary

Learn more about the ideas and references behind this note.

FAQ

What is Goodhart's Law in product design?
When a proxy measure becomes the goal, it stops being a good measure. A news app optimizing for time spent reading adds infinite scroll and clickbait. Time skyrockets but nobody is actually reading. The metric wins, the experience loses.
Why do social platforms reward outrage?
Platforms optimize for engagement, which counts any interaction equally. Rage-clicking, hate-sharing, and doomscrolling all register the same as genuine delight. The algorithm doesn't distinguish between engaged-because-happy and engaged-because-enraged.
What are some related topics to explore?
Goodhart's Lawvanity metricsdark patternsengagement optimizationproduct design ethicsdoomscrolling

Defined Terms

Goodhart's Law
The principle that once a measure becomes a target, it ceases to be a good measure.
Dark patterns
Interface choices designed to mislead users into actions that benefit the platform at their own expense.

Foundations

Over-optimization of academic publishing metrics: observing Goodhart's Law in action
GigaScience, Oxford Academic, 2019
On Goodhart's Law, with an application to value alignment
arXiv, 2024

Related Reading

Related Notes

The Data That Flatters
A soccer club spends a decade building the most expensive analytics department in its history and ends up in a relegation fight with a roster full of sprinters who cannot pass. The numbers were never the problem.
The Eternal Reinvention Tax
Designers bring dissertation-bound case studies to interviews. Engineers write 'software engineer' on LinkedIn. Why one field keeps rewriting itself.
The Part Nobody Designed For
Haptics shift how buttons feel. Colour and scent shape decisions. The emotional layer of interfaces was never in the spec but always in the experience.