Every AI pilot needs a visible not-now list

A hypothetical prioritization exercise: a small team can try only one AI workflow this month. Three proposals compete for attention: draft meeting notes, answer internal questions, or prepare client updates. A list of benefits does not yet make a decision.

For each proposal, write the intended outcome, the people affected, the review burden and a willing owner. Then name one reason to choose it now and one condition that would make you stop or defer it.

Keep a visible “not now” list for the other proposals. Explain what resource constraint drove the choice and when, if ever, the decision will be revisited. Deferral is not a promise to deliver later.

For the selected pilot, specify the decision boundary: who may change the scope, who reviews outputs and which actions still require a person. A successful draft should not silently authorize automatic sending.

Before starting, ask whether the owner actually has capacity for the review work. If nobody can take it on, reduce the scope rather than assigning responsibility by implication.

What would you put on a not-now list so that one useful experiment has a fair chance to finish?