Should an AI pilot start with a plan to scale or start with a question to learn from?
The continuous innovation panel didn't give one answer. Liz Bacelar challenged loosely framed pilots and argued for thinking through business value, approvals, policy and sensitive data before testing with real work. Amy, speaking about a global services organization, warned that demanding scale from the start can block useful learning. Both are worth holding in mind when you decide what your project needs.
Amy described small tests that can lead to expansion or to a changed plan. Sometimes the answer is stopping. A contained learning exercise and a test with live customer information shouldn't get the same treatment just because both are called pilots.
Write the learning question first, then say what the test could establish and what it couldn't. A good result from a small group doesn't prove demand across the company or readiness for every customer.
Next come the conditions for running it. Name the permitted information, the environment, the reviewers and the stopping conditions. If the test touches real operations, list the approvals you need before it starts.
Expansion is its own decision. I'd spell out the evidence and operating requirements it would need and borrow none of it from the test.
At the end, report against the original question. An honest decision to stop is a useful result. A promising one still needs an owner who can decide whether the next stage is justified.