Earlier this year, I was two days into a client training, demonstrating Microsoft Copilot's Researcher feature. The way it works, you hand it a topic, it goes off and conducts deep research on its own, and you come back a while later to a finished report. So I set it loose and the room moved on to other things.
When we came back, the report on the screen was about Pilot Purgatory.
A term straight out of the adoption framework I teach, researched and written up by someone else's AI, in front of a room I was there to train. We had a good laugh about it. But the moment stayed with me on the drive home, because it raised a question I couldn't put down: if the machines already know the term, does anyone actually know what it means?
Let's fix that.
The definition
Pilot Purgatory is the stage of AI adoption where a pilot succeeds and nothing changes. The demo impressed everyone, the test group proved the tool works, and the workflow it was supposed to transform is still running exactly the way it did before. The organization sits in between success and failure, mistaking the act of piloting for the act of adopting, and it can stay there for years.
That last part is what makes it purgatory rather than failure. A failed pilot at least produces a decision. A successful pilot that never becomes anything produces a comfortable illusion of progress, and comfortable illusions don't force decisions at all.
How you know you're standing in it
You'll recognize the sequence, because it always runs the same way.
Someone on the team found a use case. Maybe it was quote comparison, maybe meeting summaries, maybe first drafts of client emails. Leadership approved a pilot. A small group tested it for a quarter, and the results were legitimately good. People said things like "this is going to change everything." There was probably a presentation, and the presentation probably had a slide about time savings.
And then the quarter ended. The pilot group went back to their day jobs. The tool is still licensed, a few people still use it, and every few weeks someone asks in a meeting whatever happened with that AI thing. Nobody has a good answer, because nothing happened with that AI thing. The pilot didn't fail. It just never became anything.
Here's the tell, and it's uncomfortable. Ask what changed about how the work gets done on an ordinary Tuesday. Not which tools got licensed. Not what the pilot proved. What changed about the actual work, for the actual people doing it. In Pilot Purgatory, the honest answer is nothing.
Why pilots stall
The numbers on this are sobering and consistent. Gartner, IDC, RAND, and McKinsey have each studied AI project outcomes independently, and they converge on the same broad finding: the large majority of AI initiatives never produce the change they were launched to produce. You've probably also seen the most famous version of the claim, the MIT study saying 95% of AI pilots fail. I'd hold that particular number loosely, because it measures one narrow thing and gets quoted as if it measures everything. But the picture it paints matches what the other research keeps finding, and it matches what I see inside organizations every week.
I should say it matches my own history too. A while back I spent an afternoon with Zapier trying to build myself a smart agent. What I actually built was an automation that grew a little more complicated with every fix, until one day I quietly stopped opening it. The demo of each individual step worked. The thing as a whole never became part of how I worked. Pilot Purgatory doesn't need an org chart. I managed to build one at a company of one.
The mechanism behind all of this is organizational, and it is surprisingly simple.
The pilot answered a question nobody was really asking. Everyone wanted to know whether the tool works. Of course the tool works. It worked in the demo, it worked in the pilot, and it works right now in a hundred thousand other companies. That was never in doubt.
The question that actually decides adoption is a different one: will you redesign the work around what the tool makes possible? A pilot never forces that question. It runs alongside the existing workflow, in a safe little enclosure, touching nothing. The old process keeps running the entire time, which means that the moment the pilot ends, the old process is simply what remains. Software installs stall. Workflow redesigns ship.
That's also why the exit from Pilot Purgatory never comes from a better pilot, a better tool, or a bigger test group. More piloting is the purgatory.
What it looks like in an insurance agency
Here's how this plays out in the industry I spend most of my time in.
An agency pilots an AI tool for quote comparison. A producer feeds in the carrier quotes, the tool builds the comparison in minutes instead of the better part of an hour, and everyone who sees the demo is impressed. The pilot report practically writes itself.
Now look at that producer's actual Tuesday, six months later. She is still re-keying client information into the AMS by hand. Still assembling renewal packets the way she did in 2019. Still spending her afternoons on certificates and follow-up documentation instead of sitting with clients. The quoting pilot worked, and her week is untouched, because quoting was one task inside a workflow nobody redesigned. The Admin Drag actually burying her, the re-keying, the renewals, the documentation, was never on the pilot's agenda.
That agency will tell you they've adopted AI. They ran a successful pilot, after all. What they've actually adopted is a demo.
The way out
We have been here before, as it happens. When factories first replaced their steam engines with electric motors a century ago, productivity barely moved for decades. They had swapped the power source and kept the factory exactly as it was: the same central shaft, the same belts, the same floor plan designed around a machine that no longer existed. The gains only arrived when a new generation redesigned the factory itself, with a small motor at every workstation and the floor rearranged around the flow of the work instead of the location of the engine. Same technology. Completely different result. The difference was the redesign.
That is the whole story of AI adoption, told a hundred years early. The direction out of Pilot Purgatory fits inside a sentence: redesign the work first, then staff the redesigned work.
The modern data says the same thing. McKinsey's State of AI research found that high performers, the small group getting measurable value from AI, are nearly three times more likely than everyone else to have fundamentally redesigned their workflows rather than layering AI on top of existing ones. The organizations getting out of Pilot Purgatory are doing the unglamorous work of mapping where the week actually goes, naming the Admin Drag inside it, and rebuilding the workflow so the busywork lands on a governed Digital Crew instead of on their best people.
It's the same order of operations I followed when I built my own AI Chief of Staff. The value never came from the tool. It came from redefining the job the tool would hold.
Notice the sequence. The redesign comes first. The AI comes second, deployed into work that has been reshaped to receive it. Run that sequence backwards and you're piloting again, whatever you decide to call it.
Start with a diagnosis
Pilot Purgatory is stage three of five on the AI adoption curve I use with every organization I work with. It sits past the early watching-and-waiting stages, and well short of a working Digital Crew. That placement matters, because the move that gets you out of stage three is completely different from the move that gets you out of stage two, and prescribing the wrong one wastes everyone's time.
So before anyone proposes pilot number four, find out whether Pilot Purgatory is actually your stage. The AI Adoption Reality Check takes about five minutes, places your organization honestly on the curve, and names the one gap holding you back most. If the answer comes back Pilot Purgatory, the stage page will show you what the next move looks like, and you'll already know the one thing it isn't. It isn't another pilot.
Quick answers
What causes Pilot Purgatory?
Pilots test whether a tool works, which was never the real question. Adoption depends on whether the organization redesigns the workflow around the tool, and a pilot running safely alongside the existing process never forces that redesign to happen.
How long do organizations stay in Pilot Purgatory?
Years, in many cases, because a successful pilot removes the urgency a failure would create. As long as piloting feels like progress, nothing forces the harder decision to change how the work itself gets done.
How do you get out of Pilot Purgatory?
Redesign the work first, then staff the redesigned work with AI. Map where the week actually goes, name the Admin Drag inside it, and rebuild the workflow so a governed Digital Crew takes the busywork, rather than piloting yet another tool against an unchanged process.
