Capture only what helps
Ask for the request, contact preference, service area and the minimum context a reviewer needs.
Avoid restricted, unnecessary or safety-sensitive detail.
Separate urgency from routine
Define the language that triggers an immediate stop or human escalation.
Do not let an AI helper diagnose a hazard or improvise emergency advice.
Make the holding message honest
Say that the request has been recorded for review.
Do not claim a booking, attendance time, CRM update or callback schedule unless the relevant provider action genuinely occurred.
Review the evidence
Track intake completeness, escalation accuracy, reviewer corrections and follow-up ownership before expanding.
Technical evaluator notes
Evaluation should confirm architecture, data handling, knowledge ownership, permissions, human authority, QA/UAT criteria, integration boundaries, monitoring, audit evidence, release gates and rollback for the specific workflow. The guide does not claim that external systems are connected.