The cost of cutting out. The cost of carrying on.

AI projects aren’t failing because the technology doesn’t work. They are failing because the people leading the projects have started in the wrong place. 

Why the 95% really happens 

Of the million-odd projects that have started and stumbled over the last year, I can point to a million reasons why. But a few come up again and again: 

• Pilots without a business case. Teams run proof-of-concepts as “bad science experiments”, detached from real problems. 
• Automating bad processes. Instead of re-imagining workflows, organisations "AI-ify" inefficiency and just do dumb things faster.  Many companies are simply overlaying AI onto outdated workflows instead of designing processes for AI agents, not humans. The result? Amplified inefficiency that moves at digital speed, missing the transformational opportunity entirely. 
• No measurement. Leaders don’t benchmark value at the start, so they can’t prove impact later.  With Gartner predicting over $15 trillion in B2B spend flowing through AI agent exchanges by 2028, measurement isn't just good practice, it's survival. 
• Misaligned incentives. Tech teams chase novelty while the business chases savings. Neither side gets what they want. 

While many AI projects don’t create value, the problem isn’t AI. It’s us. 

Two big misunderstandings 

As I see it, there are two big misunderstandings that sit behind these failed science experiments: 

The technology. AI isn't an ERP you "install and forget". It's a continuous evolution that requires ongoing training, care, and investment in people and know-how to keep pace with its own and an organisation’s adaptation. 

The opportunity. Many organisations sit on data assets worth millions. AI can unlock new products, services, and decisive advantage, but leaders treating it as a cost-cutting exercise risk optimising their way to irrelevance.