94% of AI Projects Make Zero Money, Here's Why
I see this all the time with UK SMEs. Someone comes back from a conference, fired up about AI agents. They spin up a pilot, maybe a chatbot for customer support or an automated invoice processor. Six months later? Crickets. The pilot's still running in a corner somewhere, nobody's using it, and the CFO is asking what happened to the budget.
Sound familiar? You're not alone. In fact, you're in the overwhelming majority.
The numbers are brutal
McKinsey's latest research on enterprise AI shows that roughly 62% of organisations are experimenting with AI agents. But here's the kicker, only about 6% qualify as genuine "AI high performers", meaning AI actually contributes more than 5% to their bottom line.
That means 94% of AI adopters are failing to capture meaningful value from their investment. Not because the technology doesn't work. But because they're stuck in what researchers call "pilot purgatory", an endless cycle of experiments that never graduate to production.
The gap isn't small either. The companies getting it right see roughly 3x ROI on their deployments. The rest see a nice demo and a regrettable invoice.
Why most get stuck
There's a pattern I've seen dozens of times:
- Bolting AI onto broken processes. You automate a mess, you just get a faster mess. If your invoice workflow has seventeen manual steps and three people who "just know" what to do, slapping an AI agent on top doesn't fix it. It makes the mess slightly more expensive.
- No clear success metric. "We want to explore AI" isn't a strategy. The companies winning with this stuff measure specific things, cost per transaction, resolution time, revenue per agent deployment. Not vibes.
- Skipping the data groundwork. McKinsey's data consistently shows that infrastructure readiness is the single biggest predictor of success. If your data's scattered across spreadsheets, legacy systems, and Dave's inbox, no AI agent in the world is going to rescue you.
What the 6% do differently
The companies actually making money from AI share a few habits:
- They fix the process first, then automate. Map it, simplify it, standardise it. Then let the agent loose on something clean.
- They start with boring use cases. Not the flashy stuff. Data entry. Routing. Tier-one support. The work nobody will miss.
- They measure ruthlessly. Every deployment has a before-and-after number attached. If it doesn't move the dial, it gets killed.
Where to start if you're stuck
If you've got a pilot gathering dust, here's what I'd do this week:
- Pick one process that's repetitive, rules-based, and painful. Not your most complex one. Your most tedious one.
- Get the data clean enough. Not perfect. Good enough. 80% clean data beats 0% perfect data every time.
- Set one number you want to move. Time saved, cost reduced, errors eliminated. One metric. Track it.
Stop trying to boil the ocean. The companies winning with AI aren't the ones with the biggest budgets or the fanciest models. They're the ones who did the boring groundwork first.
References & Further Reading
Stuck in pilot purgatory?
I help UK SMEs turn AI experiments into actual results. Process first, tech second, outcomes measured from day one.
Let's have a chat