95% of AI pilot projects fail. What sets the other 5% apart.

95% of enterprise AI pilot projects never reach production with a measurable return on investment. The figure comes from a 2025 MIT study (MIT NANDA, The GenAI Divide: State of AI in Business 2025), which draws on hundreds of deployments.

After thirty years in IT, I am not surprised by the figure. I have seen this pattern before.

Several waves, the same gap

I have worked through several waves of technology: the arrival of the internet in business, the millennium bug, the digitalisation of processes, the spread of smartphones and now AI. Each time, the same gap opens between what a new technology promises and what it delivers in practice.

AI follows the same pattern. It even widens the gap, because it is easier to try and harder to scale up than anything that came before it.

The gap is one of method

The same study holds a second lesson, more useful than the first: projects handed to a specialist partner succeed about twice as often as those built in-house.

Every company has access to the same models. What sets the 5% who succeed apart is method, the subject of why AI projects fail.

What the 5% do

Companies that get a return start from their own systems and keep control at every step:

• they connect AI to the systems they already use (their ERP, their CRM), where their data and processes live;
• they move in stages, with approval points;
• they keep a human team in charge, one that owns the approach.

They treat AI as a commitment made at the very top of the company, and give it the rigour that demands.

Customer Zero: seven agents at work in Djtal

Customer Zero means being our own first customer. Seven AI agents work in our own company, organised around our internal balanced scorecard: communications, sales, operations, finance, knowledge management. Like the 5% above, each works within a written scope and answers to a supervisor.

That is how I can talk about AI with evidence to back it up.

Method can be learned

The 95% fail for lack of method, even though the technology is within reach. And method can be learned, using the systems you already have.

---

An AI project stuck at the pilot stage? Djtal works with Swiss companies on AI applied to the ERP and CRM they already run. Get in touch.