Nobody schedules a meeting to kill an AI pilot. That's the thing people misunderstand about failure rates. The pilot doesn't get cancelled, it just stops being talked about. The champion moves teams. The budget gets quietly redirected. Six months later someone asks "whatever happened to that AI thing?" and nobody quite remembers.

Industry surveys keep putting the share of enterprise AI initiatives that never reach production somewhere north of 80%. I don't quote that number to be dramatic. I quote it because, having sat in the rooms where these decisions are made, the figure matches what I see. And the reasons are remarkably consistent.

The quiet death of a pilot

Here's the pattern. A team gets excited about a tool. They build a proof of concept that demos beautifully. Leadership claps. Then the proof of concept meets reality, messy data, an unclear owner, a compliance question nobody can answer, a workflow that the tool doesn't quite fit. Progress slows. Enthusiasm cools. And because nothing formally failed, there's no post-mortem, no lesson learned, just a slow fade to nothing.

The tragedy is that the organisation concludes "AI didn't work for us, " when what actually happened is that the conditions for it to work were never put in place.

It was never a technology problem

I came to AI from the commercial side, chartered accountancy, then commercial and innovation roles across BT, Sky and ICI. So I'll admit a bias: I think most organisations dramatically overweight the technology question and underweight everything around it. The model is rarely the bottleneck. The bottleneck is the system the model has to live inside.

A brilliant model dropped into an organisation that isn't ready for it will lose to a mediocre model that is properly embedded. Every time.

When I ran the EXPLORE intrapreneurship programme at ICI Pakistan, we engaged more than 900 people and generated over 200 validated ideas. The ones that survived to deliver real money, and EXPLORE delivered over PKR 300 million in recurring annual impact, weren't the cleverest ideas. They were the ones with a clear owner, a real budget line, and a sponsor who'd absorb the early friction. The same logic governs AI today.

The four conditions that keep a pilot alive

Across the engagements I've run and watched, four conditions separate the pilots that graduate from the ones that fade. None of them are technical.

  1. A named owner with skin in the game. Not a committee. One person whose objectives are tied to the outcome, who feels the pain if it stalls and gets the credit if it ships. Pilots owned by "the innovation team" in the abstract almost always drift.
  2. A use case chosen for value, not novelty. The most impressive demo is rarely the most valuable deployment. Pick the boring, high-frequency, high-cost process, the one people do a thousand times a month, over the flashy one that wins applause but touches nothing.
  3. Governance decided up front. Who can use this, on what data, with what review? If you can't answer that on day one, the question will surface at exactly the wrong moment, usually just as you're ready to scale, and freeze everything.
  4. A path to adoption, not just deployment. Shipping the tool is the halfway point, not the finish line. If you haven't planned how people's daily behaviour changes, training, incentives, removing the old way of working, the tool becomes shelfware with a login page.

The uncomfortable truth

Three of those four conditions have nothing to do with AI. They're the same conditions that decide whether any change initiative succeeds. AI didn't rewrite the rules of organisational change, it just raised the stakes of ignoring them.

How to start differently

If you're about to launch an AI pilot, do one thing before you write a line of code or sign a tool contract: write down, on a single page, who owns it, what specific number it's meant to move, what the governance guardrails are, and how you'll get people to actually use it. If you can't fill that page, you don't have a pilot, you have an experiment looking for a sponsor.

That page is unglamorous. It's also the difference between an AI programme that compounds and one that quietly dies. The organisations winning with AI right now aren't the ones with the best models. They're the ones that took the conditions seriously before they took the technology seriously.