95% of AI projects deliver nothing. Here is what the 5% do differently

Money spent, nothing noticed

A managing director told us recently that his company had invested heavily in AI over the past year. Licences, a pilot, an enthusiastic project team. When we asked what it had delivered, it went quiet. Plenty had been tried, but in the numbers and in the workload nothing had changed.

He is not the exception. He is the rule, as the figures on AI in the Dutch SMB sector also show.

The 95 percent

A widely discussed study by MIT, Project NANDA, titled The GenAI Divide, reached a sobering conclusion: 95 percent of the organisations studied got no return at all on their AI investments, despite tens of billions in spending. Five percent won, the rest saw nothing back.

The first thing to know: that was rarely down to the technology. The models did exactly what they were supposed to do. It was down to how they were deployed.

Why the 95 percent get stuck

There are a few recurring causes, and they will probably sound familiar.

The AI was deployed in the wrong place. Many companies put it on marketing and sales, while the return sits in the back office, in the repetitive work no one misses.

The pilot never scaled to production. A pilot runs on clean data, an isolated system and goodwill. The real world adds ownership, security, integration and exceptions. Fewer than 20 percent of pilots make the leap to production. The rest remain a proof of concept forever, costing budget and delivering nothing.

There was no change management. The technology was switched on, but not woven into the daily processes. That is exactly where the gain sits, and exactly where it is most often missed.

No one measured the result. 77 percent of the companies deploying AI cannot demonstrate whether it delivers value, simply because no success criteria were agreed in advance. Without a yardstick you cannot even call a success a success.

What the 5 percent do differently

The winners share a few traits, and it is no coincidence that they do the opposite of the list above.

They focus on the back office, on well-defined processes where time saved lands directly in the numbers. They buy and integrate targeted solutions instead of building everything themselves. They choose deep, industry-specific applications instead of a general tool that fits everywhere a little. They set hard agreements on usage and governance. And they measure against concrete workflows, not against a vague sense of progress. In the MIT study the success rate of that approach was nearly twice as high as the rest.

Why pilots stall, and what does work

We wrote earlier about why AI pilots stall. The MIT study confirms it at scale: an AI that automates a task inside a process designed for humans rarely delivers impact. The gain only appears when you redesign the process around it.

That is exactly why we do not start with a tool, but with measuring. An AI audit maps in 14 days where the return really sits in your firm, which processes lend themselves to automation, and how to make the result measurable. After that we build it into the process, through targeted implementation or an AIOS layer, and not as a pilot that keeps hanging alongside daily operations.

Which side you want to be on

The difference between the 5 and the 95 percent is not a matter of budget or luck. It is a matter of where you start, what you measure and whether you truly build the technology into your process.

Fourteen days later you know exactly where the return sits for you and how to make it measurable. AI that works in your field starts not with a tool, but with the right first question.