Stuck in AI Purgatory: why your pilots never ship (and the boring fix).

"Too many enterprises are stuck in AI Purgatory — a lot of pilots and testing, but not a lot is being rolled out company-wide."

I read that line from an IT manager online, and it named the thing perfectly. Because the problem in most companies isn’t that they aren’t trying AI. They’re trying it constantly. Pilot after pilot, proof-of-concept after proof-of-concept. What they can’t seem to do is ship — get one thing into the daily workflow of a real team and leave it there, working.

The number behind the feeling

MIT’s 2025 research on enterprise AI found that 95% of AI pilots delivered no measurable P&L impact. The report is blunt about why: "the core issue isn’t the technology, it’s the learning gap." Not model quality. Not compute. The gap between a tool being available and people actually knowing how to use it for their real work.

That’s Purgatory: technically you have AI everywhere, and practically nothing has changed.

Why pilots stall

The same causes show up again and again:

• A point tool bolted onto a broken process. The demo works; the daily reality breaks on the first exception, and your best person ends up babysitting it.
• No process redesign. The workflow was never mapped, so the AI is automating chaos instead of a clean path.
• Fading sponsorship. The exec who championed it moves on; the pilot has no owner; it quietly dies.
• No learning loop. Nobody sat with the team until the tool was simply "how we do this now."

Notice none of these are technology problems. They’re adoption problems.

The boring fix

The way out of Purgatory isn’t a better pilot. It’s adoption, done in an unglamorous order:

1. Pick one workflow. One team, one painful task — not a company-wide platform.
2. Redesign the process first, then apply AI. Map it, remove the manual steps, and let the tool carry the routine volume.
3. Run it in parallel until it’s trusted, then make it the default.
4. Prove the number, then move to the next workflow.

And staff it with your own people: a small core of internal champions — roughly 2–5% of a team — plus role-based enablement so everyone else gets shown the one thing the tool does brilliantly for _their_ job.

The way out of Purgatory isn’t another pilot. It’s picking one workflow, making it stick, proving the number, and moving on.

If your organisation is running more pilots than it has shipped, that’s the signal. Start with where the waste — and the untapped capability — actually is → the free Tool Waste Self-Check.