Overview

This guide explains how to set up the "Escalate to Human" functionality in your AI agent, allowing it to notify a human operator when intervention is needed in a live chat.

Why Use Human Escalation

Human escalation is useful when:

• A user specifically requests to speak with a human
• The AI agent cannot adequately answer a question
• Certain sensitive scenarios require human judgment
• Complex situations arise that are beyond the agent's capabilities

Setting Up Human Escalation

Step 1: Create a Notification Tool

1. Navigate to the Tools section of your agent configuration

2. Select the notification channel you prefer (e.g., Slack)

3. Create a new tool named Notify_Human (or similar)

4. Add a Send Message module for your chosen platform

5. Configure the message content, for example in Slack this could be:

   🚨 Hey <@USERID>,
    A WhatsApp convo needs your attention!
   
    Name: {{ AGENT INPUT 1 }}
    Issue: {{ AGENT INPUT 2 }}
   
    Can you please take over? Go to the <https://flowgent.ai/app/chats|*live chat here*>
   

6. Add agent inputs: for example the full name and "Describe here what happend on and why you escalated this chat to a human agent"

Step 2: Update Core Instructions

Add clear guidelines in your agent's core instructions about when to escalate:

WHEN TO NOTIFY A HUMAN:
- If a user explicitly asks to speak with a human
- If you cannot answer a user's question confidently
- If the query involves [specific sensitive topics]
- If a user appears frustrated after multiple interaction attempts
- [Any other specific scenarios for your use case]

HOW TO NOTIFY A HUMAN:
1. Inform the user that you're connecting them with a human agent
2. Use the @Notify_Human tool to alert the support team
3. Respond to the user explaining that you notified a human operator to help answer the question, they will get to them asap, and if there is anything else you could help with in the mean time.

Taking Over as a Human

When responding to an escalation:

1. Lock the chat to prevent the AI from continuing to respond
2. Identify yourself as a human support agent
3. Address the user's concerns directly
4. When finished, you can either:
   • Keep the chat locked if the conversation is complete
   • Unlock the chat if you want the AI to resume handling the conversation

Best Practices

• Clear handoff messaging: Ensure users understand when they're transitioning between AI and human support
• Response time expectations: Set clear expectations for how quickly a human will respond
• Training optimization: Review escalation cases to improve your AI's capabilities over time
• Monitor escalation frequency: A high number of escalations may indicate opportunities to improve your agent's knowledge or instructions