What is the task, and who owns it?
Describe one task and the person who will use the output. Identify the task owner who can explain a good result and review the examples used to evaluate it.
What information may be used?
List the information the pilot actually needs and the permissions that apply. Discuss where information is processed and who may inspect it before choosing an implementation.
How will outputs be reviewed?
Choose representative examples and define the role of human review. A useful evaluation includes incomplete inputs, unclear requests, and cases where the system should decline to make a suggestion.
What happens when it goes wrong?
Agree who receives issues, what the fallback process is, and how access can be paused. These are operational decisions to make before the pilot is used by a wider team.
What would justify the next step?
Set a review point and the evidence needed to continue, change direction, or stop. A pilot should produce learning that informs a decision, rather than treating wider adoption as the automatic outcome.