Every case should answer five questions before anyone trusts an AI output:
- What is the user trying to get done?
- Which workflow owns it?
- Which current sources govern it?
- What can the assistant draft or recommend?
- What still requires a human decision or real-world action?
<aside>
💡
Open a populated case below and follow it through the whole loop. Use START HERE — duplicate this case when you want to try your own input.
</aside>
Three useful paths
- 20% discount request → Pricing & Approval
- Friday launch blocked by missing assets → Operations & Handoff
- Recurring executive-summary QA failure → Quality & Exception
A recommendation is not an approval. A draft is not sent. A handoff is not proof that work moved.
Untitled
Untitled