How Escalations Work
Each escalation rule has two parts:- Condition — what triggers the escalation
- Action — what happens when triggered
Conditions
Escalation conditions determine when a conversation should involve a human:| Condition | When it triggers |
|---|---|
| AI Needs Info | AI needs additional information it can’t find in the knowledge base |
| AI Missing Info | AI cannot fully answer the guest’s question |
| Message has urgent tag | Guest message matches an urgent tag |
| Message has non-urgent tag | Guest message matches a non-urgent tag (for monitoring) |
Actions
When a condition triggers, Cendra takes one of two actions:| Action | Behavior |
|---|---|
| Escalate | AI stops responding. The conversation is flagged for human review. No AI response is sent. |
| Answer and Escalate | AI sends a response first (answering what it can), then flags the conversation for human follow-up. |
Creating an Escalation Rule
- Navigate to Agent Hub → your agent → Escalations
- Click Add Escalation Rule
- Select a condition from the dropdown
- Select an action (Escalate or Answer and Escalate)
- Save the rule
Best Practices
- Start with more escalations — when you first go live, escalate frequently. Reduce as you build confidence.
- Use “Answer and Escalate” for common questions the AI can partially handle — this keeps the guest engaged while your team prepares a complete answer.
- Combine with Tags — create tags for high-priority scenarios (complaints, maintenance, refunds) and set escalation rules that trigger on those tags.
- Monitor the Dashboard — track escalation rates to identify patterns and improve your AI’s knowledge base.