Run the pilot.
Then write the policy.
If you want to run pilots you can defend, this is the guide for it. Three gates every pilot should pass, and seven ways pilots fail.
| Map | Pick one process and measure the starting point. How long does it take? How many people are involved? What are the steps? What does it cost? How often does it happen? You can't show an improvement unless you measure first. |
|---|---|
| Prove | Run the smallest version you can. One person and a subscription is often enough. Keep it small until you have a real result. |
| Scale | Only once you know it works. Two things stop it. If it costs more than paying a person and being faster doesn't buy you anything, the case isn't there. If the output isn't consistent enough to rely on, it isn't ready. A pilot that says no is successful governance. |
Three gates, three decisions, and a record of each one. That's what makes it defensible.
The guide has the rest: the seven failures and which gate catches each one, what the build actually takes, and a one-page brief you can fill in and table.
All you need to do is tell me where you're up to and where to send it. This is my gift to you.
The well-governed pilot
I guide Australian care providers who are adopting AI. I've also built AI software for this sector. So I've had my chance to learn the hard way.
Mark Scales
StartRisk