Essay · AI & Capital Discipline · Part 03 of 12
The most expensive thing I've seen a company build is a pilot that worked.
It worked. Everyone agreed it worked. The demo was good, the users liked it, the numbers in the test were real. And then eleven months later it was still a pilot, and the team that built it had been reassigned, and the vendor contract renewed on autopilot because cancelling it required a decision and nobody owned the decision.
Nothing failed. That's what makes it hard to catch.
Success is the dangerous outcome
A pilot that fails gets killed. The failure is loud, the budget comes back, everyone moves on. Failure is cheap and self-cleaning.
A pilot that succeeds enters limbo. It has evidence on its side, so nobody can argue against it. It has no production owner, so nobody can advance it. It sits there generating a small recurring cost and a large amount of internal goodwill, and the goodwill is what keeps it alive.
I've come to think of this as the real failure mode of corporate innovation. Not the projects that don't work. The ones that work and have nowhere to go.
The question that prevents it
Before a pilot starts, someone has to answer this: who owns this in production, and what does it cost them?
Both halves matter. The first half is usually easy to answer badly. Someone volunteers, or a name gets written down, and everyone feels covered. The second half is where it gets real. Production ownership means headcount, budget line, an on-call rotation, a support burden, a place in someone's operating plan.
If no one has agreed to absorb that cost, the pilot has no destination. You can still run it, and sometimes you should. But you're running an experiment, not a rollout, and you should say so out loud before you start rather than discovering it afterwards.
What I got wrong about this
For a long time I thought the fix was better handoffs. Cleaner documentation, longer overlap between the pilot team and the operating team, more thorough runbooks.
It isn't. Handoff quality barely moves the outcome. What moves it is whether the receiving team helped set the success criteria in the first place. A team that had a say in what "working" means will take the thing. A team handed a finished product they had no voice in will find reasons it doesn't fit their environment, and they'll usually be right, because it doesn't.
So the operating owner isn't a recipient. They're a participant from the first week, or the handoff is going to fail no matter how good the documentation is.
Kill criteria, written down before
The other half of ownership is permission to stop. Write down, before the pilot begins, what result would end it. Not vague language. A threshold and a date.
This feels unnecessary while everyone is optimistic. It's the only thing that works later, when the pilot is ambiguous and the team is attached to it and the person who'd have to call it is the person who championed it. Nobody wins that argument on the day. You win it in advance or not at all.
On Monday
Pull the list of live pilots. For each one, name the production owner and the budget line that will carry it. Then name the result that would kill it and the date you'll check.
Any pilot missing either answer isn't on a path to production. Either give it one this week or stop paying for it.
The operating principle. A pilot without a named production owner and a written kill threshold isn't a pilot. It's a subscription.
Juan Vegarra is the author of An Outsider’s Playbook. The views here are his own. More essays · Advisory · Write me