There's a specific kind of AI project death that looks nothing like a failure. The pilot worked, the users liked it, the champion is enthusiastic — and after ten months, no contract has been signed, the pilot is still "running," and the team is slowly feeding it real work. Welcome to pilot purgatory: not canceled, not purchased, living in the space between demo and decision. It's where most applied-AI opportunities quietly die, and both sides usually know it.
Why pilots get stuck in the first place
The pilot is the safest procurement device ever invented. It lets everyone say yes without anyone deciding anything. The vendor gets access and a story. The innovation team gets a project that isn't an expense line item. The IT department isn't committing to a tool. The CFO isn't signing a budget. Nobody is accountable for an outcome, because it's "just a pilot."
The structural reason pilots stall is that the pilot answerable to nobody. The decision the pilot was supposed to produce — "is this worth production money, and who pays for it?" — was never assigned a moment, a person, or a budget line. So the pilot keeps producing evidence while the organization keeps not deciding.
How to tell you're in purgatory
The tell-tale signs are consistent, and once you see them you can't unsee them:
- No exit criteria exist. Nobody can tell you what evidence would end the pilot — either with a contract or with a polite thank-you.
- There is no end date. The pilot runs on "let's keep validating."
- The champion has enthusiasm but no budget. Their power extends to "try it," not to "buy it."
- You keep talking to the innovation team. The decision lives in the business unit, which you haven't met.
- The pilot is secretly production. It's handling real work, in business hours, and people depend on it — while it's still officially an experiment.
If three of these are true, you're not in a sales cycle. You're a free dependency of an organization that would collapse into a decision if you stopped showing up. Which, by the way, is the one move that reliably breaks purgatory — but only if you do it with evidence on the table.
How to structure a pilot that converts
The fix is upstream: structure the pilot so the decision already exists the day it starts.
- Define the decision in the kickoff. Schedule the meeting where the pilot results are reviewed. Put a specific person in charge — the one who owns the budget, not the one who owns the enthusiasm.
- Timebox it. Six weeks is usually enough. If success criteria can't be met in six weeks of real work, they won't be met in six months either — the project is too big, or the criteria are too vague.
- Get the budget conversation started before the pilot does. Not a signature — a line item. "This replaces a workflow that costs X; here's where the money sits." If there is no money anywhere, the pilot was never going to have a destination.
- Charge for the pilot. Even a token amount. It's not about the revenue; it's about what happens in the buyer's head the moment there's an invoice. Free pilots silently teach an organization that the tool is free.
- Make the pilot painful to ignore. Real users, real workflows, a written report every two weeks answering one question: "here's what was supposed to happen, here's what happened, here's the measured difference." An evidence trail that the champion can drop into the budget meeting.
When to walk away
The other side of the same discipline: purgatory is a choice, and you're allowed to choose differently. When the pilot hits its timebox without a decision, the healthiest move is often to end it yourself — with the evidence report, a fair price, and a standing offer. The teams I've seen do this consistently gained two things: weeks of their own time back, and a reputation for seriousness that beats any discount. The buyer who wasn't going to buy hears you name the price, calmly, and realizes the offer actually expires.
Vendors get stuck because they treat the pilot like a funnel stage. It's a relationship contract. The moment it stops having an end, it never finds one on its own. Put the expiry date on the pilot before you start it, be the one who notices when it expires — and be willing to be the one who calls it.
An organization that can't decide after six weeks of evidence wasn't going to decide after six months. It was going to run out of money, or a champion, or both — and take your time with it.