"We need a presentation" describes a deliverable. "We need to decide which projects to continue" describes a purpose. When you define the expected outcome of an AI initiative, connect what will be produced to what that output should make possible. That link guides the format, the review and the moment the work has gotten where it needed to go.
Write down the deliverable and its reason
Start with two sentences: I want AI to help with a certain piece of work; I am asking for it because a certain situation needs to move forward. Then check whether there is a clear link between them.
A team asks for a dashboard to track requests. The reason might be to find requests with no owner. Then the dashboard has to make that gap visible; lots of volume charts may not help. The exercise may show that a reviewed list is enough for the decision at hand.
Make the outcome observable
"Improve management" is too broad. Describe a change the team can recognize: finding requests with no owner before the meeting, for example. Also define the conditions: which sources, which part depends on confirmation and who reviews the deliverable. This keeps you from promising more than the experiment can show.
Go back to the agreement after the first version
Compare what was produced with the original reason. Does the deliverable let you do what motivated the work? If information or an action is still missing, write down that gap. A clear statement does not stop you from learning along the way; it gives you a reference for recognizing what changed.
In AI Thinking by Dani Hayoshi, the Question, Redesign and Rebuild moves explore exactly this relationship between need and path. See also AI for leaders.
Dani Hayoshi is the author of AI Thinking. Her work connects artificial intelligence, education and entrepreneurship. About Dani.