The pilot that never leaves the pilot

Boardroom. Month nine. On the wall: “AI pilot. Status: piloting.” A tray of untouched sandwiches.

The general manager says there is great energy in the room and asks where we are. IT says security has not signed off. Marketing says it wrote some great drafts. Finance asks what it cost. Operations thought Sam owned it. Sam, at the end of the table, is in sales.

If that meeting sounds familiar, you are not alone. When Orgvue surveyed 1,100 leaders at medium and large organisations in May 2026, 43 percent said they had AI projects stuck in pilot and 35 percent reported outright failures. Every number like that describes the same meeting. A tool was bought. A champion tried it. Nobody’s name was next to the outcome.

It is almost never the model

Pilots do not stall because the AI could not do the work. They stall because the work was never defined well enough for anyone to say it was done. “Explore how AI could help with quoting” cannot fail and cannot succeed. It can only continue.

Three things are missing in nearly every stalled pilot, and they are the same three we found missing in our own business before we reset it in August:

1. An outcome, in the business’s own numbers. Quotes out same day. Tickets closed remotely. Days sales outstanding. Not “adoption”.
2. An owner, by name, who signs when the outcome is reached and is answerable when it is not.
3. A cost, logged per week, so Finance never has to ask the question in month nine.

Why mid-sized businesses get stuck harder

A ten-person business can run a pilot on the owner’s laptop and kill it on a Friday. A hundred-person business has an IT manager who, quite rightly, needs five things in writing before anything touches the tenant: where inference runs, who the sub-processors are, who approved what, that nothing can be set to “always allow”, and that the thing runs under the company’s own identity and logging.

When nobody owns the outcome, nobody produces those five things, and the pilot sits in the gap between “security hasn’t signed off” and “it wrote some great drafts” indefinitely. Governance is not the brake. It is the thing that lets you drive. But somebody has to write it, and that somebody needs a name.

What the one that shipped looked like

Same company, month ten. The screen now reads:

One outcome: quotes out same day. Owner: Priya. Done when: 90 percent within 24 hours, measured from the job system. Cost: logged per week.

Priya has her hand up. The sandwiches are being eaten. Nothing about the technology changed between the two meetings. What changed is that the pilot became a job, and a job has a definition of done.

The pilot-rescue review

For businesses between twenty and two hundred staff, this is the first thing we offer: three weeks to establish what was bought, what it is actually doing, what one outcome to ship first, and which of the five governance items are missing. It ends with a named owner, a definition of done and a fixed quote for the implementation, in your tenant, under your identity and logging, with our supervision on top.

The missing line item is a name. That is what an AI operations partner is. Tell us where the pilot is stuck and we will tell you, plainly, what it would take to get it out.