The Best AI Projects Did Less.
The AI projects that worked this year were embarrassingly small, and the ones that died were the impressive ones. That is the finding nobody wants on the slide, so we will print it.
MIT's NANDA study put a hard number on it: 95% of enterprise generative AI pilots produced no measurable impact on the profit and loss statement. And the lead researcher said the quiet part out loud: model quality is rarely the constraint.
So read the room. The default reaction to a 95% failure rate is to assume the survivors had something better. Wrong. They did something smaller. The pattern in the 5% that worked is not ambition. It is restraint.
The grand pilots die on contact with a Tuesday: the messy data, the undocumented exception, the handoff between two teams that hate each other. The boring survivor does one repeatable task, watches its one seam, and quietly gets more reliable every week.
If your AI project is not moving the numbers, the answer is almost certainly not more model, more scope, or more budget. It is less, aimed better, and remembered so it stays reliable.
The demo told you to dream bigger. The data is telling you to ship smaller.
Full read on the blog.