Skip to main content

Playbook: AI to Human Handoff - Using Intents for Smooth Customer Escalation

Updated this week

🔍 What Are Intents?

Intents are powerful conversation triggers that detect when a customer expresses a specific need or interest — such as wanting a discount, asking about a product, or being ready to purchase. When an intent is caught, it can automatically trigger a flow, helping you guide the customer towards conversion.

🎯 Goal of This Playbook

Use Human Escalation Intents to improve your customer satisfaction by:

  • Detecting escalation signals (e.g., requests for human help or expressions of frustration)

  • Triggering an automated handoff flow that acknowledges the request and alerts your support team

  • Ensuring seamless transition to human agents with full conversation context

  • Boosting both customer trust and support team efficiency


💡 Use Case: Escalate to a Human Agent

Customers often ask:

  • "Can I speak to a real human"

  • "This doesn’t quite solve my problem yet"

  • "No, this wouldn’t work for me"

These are signals where we’d need to break out of the AI conversation and have a human agent take over.

With Intents, you can catch these questions and respond in a way that escalates the requests accurately and timely.


🔧 Step-by-Step Setup

Creating Intents

  1. Start by setting up the intent. For this, navigate to the “Intents” page and click create new.

  2. Next up, define a generic sentence that represents your intent, e.g. “Can I speak to a human”, “This doesn’t work for me”.

  3. Use the “Generate samples” button to let our AI help you create more example phrases, such as:

    1. Could you connect me with a human representative?

    2. I need actual human help

    3. I want to reach customer service directly

    4. This doesn’t solve my problem

    5. This is not helpful at all

  4. Edit them as you like, the more phrases, the more complete the intent

  5. Click Save if you are happy with your intent

Creating the flow

  1. Use the intent you just created as the trigger for your flow

  2. Send a message that indicates the support team will take over

  3. Set “hide flow messages in feed” to false, so that the feed will be marked as unread in the conversations screen, for the customers’ human agents to find the conversation and reply.


📈 Benefits

  • Better customer satisfaction: Customers feel heard when their requests for human help are recognized and acted upon promptly

  • Reduced customer frustration: Instead of customers getting stuck in automated loops, they're quickly routed to human support when needed

  • Prevents conversation abandonment: Catching escalation signals early prevents customers from leaving the conversation frustrated

  • Timely intervention: Automated detection ensures no escalation requests are missed or delayed

  • Seamless handoff experience: Customers get immediate acknowledgment that their request for human help has been understood and will be addressed

  • Improved agent efficiency: Support teams can easily identify conversations that need human intervention through unread feed notifications

  • Maintains conversation context: The entire conversation history is preserved for human agents to understand the customer's journey


✅ Best Practices

  • Capture comprehensive escalation language: Use AI-enhanced generation to include formal, informal, frustrated, and polite ways customers request human help

  • Include negative sentiment indicators: Add phrases expressing dissatisfaction or indicating automated solutions aren't working for their specific needs

  • Monitor and optimize regularly: Track escalation patterns to identify automation gaps and test the handoff process for smooth agent takeover


🔚 Wrap-Up

By setting up human escalation intents, you're building trust with customers who know they can always reach a real person when needed, while streamlining your support workflow. It's a balance between automation efficiency and human touch that keeps customers happy and your team focused on high-value interactions.

Did this answer your question?