Your safety net for automated conversations
The Escalation Settings feature acts as a powerful safety net, ensuring that customers who need human attention are seamlessly handed over from an AI agent. This prevents frustrating loops and ensures complex or sensitive issues are handled by the right person.
You can find these new settings in a new "Escalation" tab within your agent's configuration screen.
Note: All escalation triggers will be disabled by default. This ensures your existing agent setups are not affected until you choose to configure them.
1. Understanding Escalation Triggers
You can choose from several predefined triggers to automatically escalate a conversation.
Based on negative sentiment
This trigger monitors the customer's messages for negative language or tone. If a negative sentiment is detected and persists for a defined period, the conversation is automatically escalated. This is ideal for catching frustrated customers early.
After getting stuck
This trigger activates when the AI agent fails to answer a question or the conversation goes off-topic after a set number of attempts. For example, if the FAQ agent cannot find a relevant answer twice in a row, the conversation can be escalated to a human who can help.
In custom scenarios
This is the most flexible and powerful trigger. You can use natural language to define specific scenarios that should lead to an escalation. This allows you to create rules for countless situations, such as:
When a user explicitly asks to speak to a human (e.g., "talk to a person," "I need customer support").
When a user mentions a specific, urgent keyword (e.g., "complaint," "broken item").
This trigger will eventually replace many of the complex intent-based flows used for escalations, offering a simpler and more powerful alternative.
2. Configuring Escalation Actions
Once a trigger condition is met, you can define what happens next.
Assign to a human agent
You can specify which human agent or team the conversation should be assigned to. This ensures the escalated ticket appears in the correct person's queue, ready for them to take over.
Customize the handover message
You can instruct the agent on what to say when it escalates the conversation. This is not a fixed message but a prompt for the AI. You can keep it simple or provide specific instructions.
Examples:
"Say you're sorry, and a human will take over."
"Acknowledge their request and let them know you're connecting them to the support team."
3. Troubleshooting
If escalations are not working as expected: Double-check your trigger conditions. The natural language "custom scenarios" may need to be iterated on to get the phrasing just right.
To return to default behavior: If you encounter any issues, you can simply disable all triggers in the Escalation tab. This will stop all automatic escalations and revert the agent to its standard behavior.