How to Create an AI Bot with Human Handoff

The hardest part of any ai chatbot human handoff escalation system is knowing when to stop and pass to a human. Get it wrong in either direction and you lose. Too early and you waste agent time. Too late and you frustrate the customer. I build these with three detection layers: content-based, sentiment-based, and conversation-length triggers.

The goal is a smooth transition where the human agent gets full context and the customer never has to repeat themselves.

What You Need Before Starting

• A working AI chatbot (see system 373)
• A ticketing or live chat system (Zendesk, Intercom, or a custom queue)
• Python 3.8+ with the Anthropic SDK
• Slack or your team chat for agent notifications

Step 1: Define Handoff Triggers

Build a detection system that catches multiple escalation signals:

HANDOFF_KEYWORDS = [
    "speak to a human", "talk to someone", "real person",
    "manager", "supervisor", "complaint", "lawyer", "cancel"
]

def check_keyword_trigger(message):
    message_lower = message.lower()
    return any(keyword in message_lower for keyword in HANDOFF_KEYWORDS)

def check_frustration_trigger(messages):
    """Trigger if user sends 3+ messages with negative sentiment."""
    recent = messages[-3:]
    frustrated_count = sum(1 for m in recent if is_frustrated(m["content"]))
    return frustrated_count >= 2

def check_loop_trigger(messages):
    """Trigger if bot has responded 5+ times without resolution."""
    bot_messages = [m for m in messages if m["role"] == "assistant"]
    return len(bot_messages) >= 5

Step 2: Add Sentiment Detection

Use the AI model itself to detect frustration:

def is_frustrated(text):
    response = client.messages.create(
        model="claude-sonnet-4-20250514",
        max_tokens=10,
        messages=[{
            "role": "user",
            "content": f"Is this message frustrated or angry? Reply only YES or NO.\n\nMessage: {text}"
        }]
    )
    return "yes" in response.content[0].text.lower()

Step 3: Execute the Handoff

When any trigger fires, transition the conversation:

def execute_handoff(session_id, conversation, trigger_reason):
    # Generate a conversation summary for the agent
    summary = generate_summary(conversation)

    # Create a ticket in your support system
    ticket = create_ticket(
        session_id=session_id,
        summary=summary,
        full_conversation=conversation,
        trigger=trigger_reason
    )

    # Notify the agent
    notify_agent(ticket)

    # Send the customer a handoff message
    handoff_message = "I am connecting you with a team member who can help further. They will have our full conversation so you will not need to repeat anything."
    return handoff_message

Step 4: Generate Agent Context

Give the agent everything they need to pick up seamlessly:

def generate_summary(conversation):
    messages_text = "\n".join([f"{m['role']}: {m['content']}" for m in conversation])

    response = client.messages.create(
        model="claude-sonnet-4-20250514",
        max_tokens=300,
        messages=[{
            "role": "user",
            "content": f"""Summarize this support conversation in 3 bullet points:
1. What the customer wants
2. What has been tried
3. Why it needs a human

Conversation:
{messages_text}"""
        }]
    )
    return response.content[0].text

Step 5: Handle the Agent Takeover

When the agent starts responding, the bot stops:

def handle_incoming_message(session_id, text, sender_type):
    session = get_session(session_id)

    if session["status"] == "handed_off":
        if sender_type == "agent":
            relay_to_customer(session_id, text)
        else:
            relay_to_agent(session_id, text)
        return

    # Normal bot flow
    response = process_with_ai(session_id, text)

    if should_handoff(session_id, text):
        execute_handoff(session_id, get_history(session_id), "trigger_detected")
        update_session_status(session_id, "handed_off")
    else:
        send_to_customer(session_id, response)

What to Build Next

Track handoff rates and reasons. If a specific topic causes most handoffs, add that knowledge to the bot. The goal is to reduce handoff rate month over month while keeping customer satisfaction high.

Related Reading

• The Handoff Problem - the deeper challenge of AI-to-human transitions
• AI in Customer Service - where handoff fits in the broader support strategy
• The Measurement Framework That Actually Works - how to measure handoff quality and timing