Pular para o conteúdo
SynapsAI Docs

Operator Handoff

Este conteúdo não está disponível em sua língua ainda.

Operator handoff is the process of transferring a conversation from the AI agent to a human operator. When configured correctly, the transition feels seamless to the user — they stay in the same chat window and simply start talking to a real person.

When does handoff happen?

Section titled “When does handoff happen?”

Handoff can be triggered in two ways:

1. User-initiated

Section titled “1. User-initiated”

The user explicitly asks to speak with a human. Common phrases include:

  • “I want to talk to a person”
  • “Can I speak to a real agent?”
  • “Transfer me to support”

The AI agent detects this intent and initiates the handoff automatically.

2. AI-initiated

Section titled “2. AI-initiated”

The agent determines it cannot adequately help the user and proactively offers to transfer. This typically happens when:

  • The question falls outside the knowledge base and the agent has no relevant answer.
  • The user has asked the same question multiple times without satisfaction.
  • The conversation involves a sensitive topic the agent is instructed not to handle (configured in the system prompt).
  • A trigger rule fires based on specific conditions.

How handoff works

Section titled “How handoff works”
  1. Trigger — the handoff is initiated (by user request or AI decision).
  2. Notification — online operators receive a notification that a conversation needs attention.
  3. Assignment — an operator accepts the conversation and takes over.
  4. Live chat — the operator responds directly to the user. The user sees the messages in the same chat window.
  5. Resolution — once the issue is resolved, the operator can hand the conversation back to the AI or close it.

During handoff, the operator sees the full conversation history, so the user doesn’t need to repeat themselves.

Configuring handoff behavior

Section titled “Configuring handoff behavior”

System prompt

Section titled “System prompt”

The most effective way to control handoff is through the system prompt. Add instructions like:

If the user asks to speak with a human, transfer the conversation to an operator immediately. If you cannot answer a question after two attempts, offer to connect the user with a support agent.

Triggers

Section titled “Triggers”

Use Triggers to automate handoff based on conditions:

  • Keyword match — transfer when the user mentions specific words (e.g., “cancel”, “refund”, “complaint”).
  • Sentiment detection — escalate when the user appears frustrated.
  • Unanswered questions — trigger after the agent fails to provide a relevant answer a set number of times.

Fallback message

Section titled “Fallback message”

Configure a fallback message that the agent sends when initiating handoff:

“Let me connect you with a team member who can help. Please hold on for a moment.”

This sets the user’s expectation while an operator picks up the conversation.

What the user sees

Section titled “What the user sees”

The handoff experience is designed to be transparent:

  1. The agent sends a transition message (e.g., “Connecting you with a team member…”).
  2. The operator joins and introduces themselves.
  3. The conversation continues in the same chat window — no new tabs, no phone calls.
  4. When the operator hands back to the AI, the agent resumes naturally.

What the operator sees

Section titled “What the operator sees”

When a handoff occurs, the operator’s dashboard shows:

  • A notification highlighting the new conversation.
  • The full conversation history — every message between the user and the AI.
  • User context — channel (web, WhatsApp), language, and any metadata.
  • Internal notes from other operators (if any).

The operator can type responses directly, which appear to the user as live chat messages.

Handling offline scenarios

Section titled “Handling offline scenarios”

If no operators are online when a handoff is triggered:

  • The agent can inform the user: “Our team is currently offline. We’ll get back to you as soon as possible.”
  • The conversation is flagged for follow-up when an operator comes online.
  • Optionally, configure a fallback action — like collecting the user’s email or phone number for a callback.

Best practices

Section titled “Best practices”
  • Always have a fallback — don’t leave users hanging if no operators are available. Collect contact info or set expected response times.
  • Keep handoff fast — users expect a human to respond within 1-2 minutes after the transfer. Staff appropriately during peak hours.
  • Train the AI to hand off early — it’s better to transfer proactively than to frustrate the user with unhelpful answers. Use your system prompt to set clear boundaries.
  • Review handoff conversations — check Conversations regularly to understand why handoffs happen. If the same topic triggers frequent handoffs, improve the knowledge base to reduce them.
  • Use internal notes — when an operator resolves a case, they should leave notes so the next person (or the AI) has context if the user returns.

Next steps

Section titled “Next steps”