Pilot purgatory isn't bad luck. It's what happens when AI activity grows faster than the organisation's ability to absorb it. Here are six symptoms, what people say when they have them, and what to do instead. There's a self-check at the end. Be honest; nobody's watching.
There is a special place where good AI ideas go to live forever without ever growing up. I call it pilot purgatory. It's well funded, full of enthusiastic people, and it has excellent demo days.
Nobody plans to end up there. You get there one sensible decision at a time: a pilot here, a proof of concept there, a team that "just wants to try something". Each one makes sense on its own. Together, they add up to a lot of activity and remarkably little change.
The cause is almost never the technology. It's fragmentation: the pieces of an AI transformation moving at different speeds, in different directions, owned by different people. Or by nobody.
The six signs
01Vision without execution
“We have an AI strategy. It's on SharePoint.”
The strategy exists, beautifully formatted, and has not influenced a single budget decision, objective or team priority. The organisation has an AI ambition, but no translation of it into the work people actually do on Monday.
Pick three business outcomes for the next twelve months, give each a named owner, and make AI one of the means, not the headline.
02Data without trust
“The model's great. Finance still exports it to Excel to check.”
The data is there. People just don't believe it, or the things built on top of it. So every AI output gets a human double-check, which quietly doubles the work instead of halving it.
Ask the people who use the output what would make them trust it. Fix that first. It's usually ownership and definitions, not technology.
03Workflows never redesigned
“We added AI to the process.”
Same process, now with an extra step. AI has been bolted onto how work was done in 2019, and everyone is surprised the gains are smaller than the vendor's slide promised.
Map one end-to-end workflow. Decide where AI informs, where it drafts and where it acts. Then remove the steps that no longer make sense.
04Governance added too late
“Legal will have a quick look before go-live.”
Narrator: it was not a quick look. Risk, ethics and EU AI Act questions arrive after the build, when they're expensive to answer and easy to say no to.
Bring risk and compliance in at the idea stage, with lightweight guardrails that tell teams what they can do, not only what they can't.
05Pilots without scaling logic
“We've got dozens of pilots running!”
Said with pride, which is the giveaway. Nobody decided upfront what success looks like, who pays for scaling, or which pilots get stopped. So none of them stop, and none of them scale.
Before any pilot starts, write down the business result it must show, the date you'll decide, and who funds the next step if it works.
06No clear ownership
“Everyone's very aligned on this.”
Everyone is involved. Nobody is accountable. The steering committee steers, the centre of excellence excels, and the outcome belongs to no one in particular.
For every AI-enabled outcome, name one business owner. One. If two people raise their hands, you have zero owners.
Tick one answer per row. Your result lights up below.
| Sign | Not us | A bit | Painfully us |
|---|---|---|---|
| Vision without execution | |||
| Data without trust | |||
| Workflows never redesigned | |||
| Governance added too late | |||
| Pilots without scaling logic | |||
| No clear ownership |
The way out
Here's the good news: fixing fragmentation rarely means starting again. It means connecting what's already there. One shared picture of where you stand. A short list of outcomes that matter. Owners who can say no. And the discipline to stop the pilots that have had their moment in the sun.
As one of our training participants put it: “The exchange with other participants confirmed that these challenges are not unique to our company.” They never are. Which is oddly reassuring, and also no excuse.
