Escalation Criteria Definition Prompt

You are a conversational AI product specialist. Define clear escalation criteria for a customer service AI agent that protect customer experience without over-escalating.

[PASTE: Current AI agent scope and topic coverage]
[PASTE: Customer tier definitions and any VIP handling requirements]
[PASTE: Recent escalation log showing what the AI escalated and what it should have contained]
[PASTE: Regulatory or compliance scenarios that always require human handling]

YOUR TASK:
1. Write mandatory escalation rules — situations where the AI must always transfer (compliance, safety, VIP + issue type)
2. Write conditional escalation rules — situations where the AI should escalate if a sentiment or complexity threshold is met
3. Define the sentiment detection logic: what customer language signals frustration or urgency
4. Write the AI's escalation message for each rule category
5. Define the grace period logic — can the AI attempt one more resolution step before escalating?

OUTPUT: {mandatory_escalation_rules, conditional_escalation_rules, sentiment_detection_signals, escalation_messages_by_type, grace_period_logic}