Why 95% of enterprise AI spend shows zero return
In 1975, economist Charles Goodhart wrote a line that explains almost every failed tech rollout in corporate history: “When a measure becomes a target, it ceases to be a good measure.”
If you look at why so many massive enterprise AI investments are failing to show real financial return right now, Goodhart’s Law is the exact culprit.
Most organizations approach AI with a genuine goal: build better products, solve harder problems, and operate smarter. Because real transformation takes months of messy experimentation to show up on a balance sheet, leadership defaults to measuring what is easy to count today.
The moment those easy metrics become the target, the incentives break.
1. Measuring speed instead of direction
When “output velocity” becomes the primary KPI, teams naturally optimize for volume. Engineers ship three times as many pull requests; writers publish five times as many drafts; analysts generate ten times as many summaries.
On a quarterly slide, productivity looks like a hockey stick. In reality, you often just end up with ten times more noise to review, maintain, and untangle. Driving twice as fast doesn’t help if nobody stopped to check the map.
2. Optimizing away the thinking time
There is a dangerous assumption in modern tech that any minute spent wrestling with a problem manually is wasted friction.
In reality, the slow, unglamorous work of digging through raw inputs, hitting dead ends, and debating ideas with peers is how people develop intuition. When you automate the entire process just to shave forty minutes off a task, you get an instant answer, but you strip away the deep understanding required to know whether that answer is actually right.
3. Tooling over problem-solving
Anyone who has spent time building with modern AI tools knows how easy it is to fall into the trap of over-engineering the setup. Chaining together endless integrations, automating trivial workflows, and celebrating the complexity of the system feels productive, even when the end result isn’t actually useful.
The builders who actually win
The people and companies getting real value out of AI aren’t treating it as a shortcut to do the exact same routine work in a fraction of the time. They are using the time AI frees up to tackle problems they previously couldn’t afford to touch: testing wilder ideas, building deeper prototypes, and spending more energy on human judgment and taste.
AI amplifies whatever system you plug it into. If the target is just “more, faster,” Goodhart’s Law guarantees you will hit the target and miss the point.