Escalation Rules

Smart escalation is what separates good AI support from frustrating chatbot experiences. Well-configured automation rules ensure customers reach a human when they need to — and only when they need to.

Types of Escalation Triggers

Confidence-Based The AI escalates when its confidence score falls below your threshold. This is the default and handles most cases.

Keyword-Based Specific words or phrases trigger immediate escalation:

"Cancel my account"
"Speak to a manager"
"Legal" or "lawsuit"
Profanity or aggressive language

Sentiment-Based The AI detects negative sentiment and escalates before the customer explicitly asks. Configure sensitivity levels for your comfort.

Topic-Based Certain topics always require human handling:

Refund requests over a specified amount
Security incidents
Enterprise contract negotiations
Complaints

Routing Rules

Escalation is only half the equation. Routing determines who receives the conversation:

Skill-Based Routing Route to agents with relevant expertise:

Billing issues go to the billing team via ticketing
Technical issues go to engineering support
Enterprise inquiries go to the sales team

Load-Based Routing Distribute conversations evenly across available agents. Prevents one agent from being overwhelmed.

Priority Routing VIP customers, enterprise accounts, or urgent issues jump to the front of the queue.

Configuration in Corebee

Go to Settings > Automation > Escalation Rules
Create rules using the visual builder or JSON editor
Set priority order — rules are evaluated top to bottom
Test with sample conversations in the playground

Best Practices

Start with 5-10 rules and expand as needed
Review escalation logs weekly to catch over-triggers
Include context handoff — the agent should see the full AI conversation and customer details
Set SLA timers for escalated conversations
Always have a fallback rule for unmatched escalations

Next up: Building custom actions that let the AI take real actions on behalf of customers.