Setting Up Escalation to Human Agents

5 min read Updated Nov 3, 2026 AI Chatbots

Escalation is the moment your AI chatbot hands a conversation to a person - and it is where chatbot experiences are won or lost. Done right, the bot absorbs routine questions and quietly brings in your team exactly when a human adds value. You configure it in two places in the AI Agent tab: When the AI responds in Agent settings, and the Handoff section.

How Handoff Works

  1. The bot detects an escalation trigger - the visitor types one of your trigger phrases, the bot cannot answer, or an AI Action escalates the chat.
  2. The bot sends your escalation response and your team is notified in the Inbox and your connected platform channel (Teams, Slack, Google Chat, Zoom, or Webex).
  3. An agent replies and takes over; the full bot conversation is visible so the visitor never repeats themselves.
  4. If nobody is available, the offline flow captures the visitor's contact details instead. See What Happens When Everyone Is Offline.
Handoff settings in the AI Agent tab showing escalation options

Configuring Handoff

Choose When the AI Responds

Go to Chat Widgets > your widget > AI Agent > Agent settings and pick an option under When the AI responds. AI only lets the AI answer every chat and notifies your team only when a visitor asks for a person. AI + human agents, AI drops when an agent answers adds both and steps the AI aside as soon as an agent replies. AI answers missed chats, AI answers when your team is offline and AI answers offline or missed chats give your team the first chance.

Open Handoff

Click Handoff in the AI Agent tab's left menu.

Set the Trigger Phrases

Under Human escalation, enter comma-separated Trigger phrases that hand the chat to your team, such as talk to a human, agent, representative. Wildcards work: *agent* matches the word anywhere in a message. Then instruct the bot (in your system instructions) to offer a handoff whenever it cannot answer from knowledge or the visitor sounds frustrated. Belt and suspenders: trigger phrases catch explicit requests; the instructions catch judgment calls.

Set the Escalation Response

The Escalation response is what the visitor sees while your team is pinged: "Let me connect you with the team - someone will be right with you." Keep it honest about wait expectations.

Handle "No Answer" Replies

Under When the AI has no answer, add No-match trigger phrases that signal the AI could not answer (for example *I don't know*, *I'm not sure*), a No-match response sent instead of that reply, and No-match quick replies such as "Talk to a human".

Route the Escalation

Escalations notify your Inbox and the channel the widget routes to. To send different topics to different teams - sales versus support, or by product line - use an AI Action with the action type Escalate chat to human (Custom Actions are included on Pro and Business plans and during your free trial). See Escalation Routing with AI Actions. Click Save changes.

Test the Whole Path

From the live widget (not just the Playground), type "I want to talk to a human." Confirm the escalation response appears, the notification lands in your channel, and an agent reply reaches the visitor. Test the offline path too by going offline first.

Tuning: Not Too Eager, Not Too Stubborn

  • Bot escalates too easily - Visitors say "agent" out of habit and get handed off before the bot even tries. Fix: use more specific trigger phrases, and have the bot offer one useful answer first: "I can likely help with that - what's your question? Or I can connect you with the team now."
  • Bot resists too long - Nothing angers visitors like a bot that won't give up. Fix: escalate immediately on the second request for a human, on billing disputes, and on anything emotional.
  • Escalations at 3 a.m. - Pair handoff with your online schedule so after-hours escalations become lead captures instead of silent waits.

Measuring Escalation Health

In Analyze, watch your escalation rate - the share of bot conversations handed to humans. There is no universal "right" number, but the trend is diagnostic: a rising rate means the knowledge base is falling behind your visitors' questions (close the gaps); a rate near zero on high volume may mean the bot is stonewalling people who wanted help.

Frequently Asked Questions

Does the agent see what the bot said?

Yes - the entire conversation, bot messages included, is visible in the Inbox and the platform thread, so agents pick up with full context.

Does the AI keep answering after an agent joins?

It depends on When the AI responds. With AI + human agents, AI drops when an agent answers or AI answers offline or missed chats, the AI steps aside once a human replies. With AI + human agents on all chats, both stay in the chat.

Can the bot escalate to a specific person?

Escalations go to channels/teams rather than individuals, and any available agent takes over. With escalation routing you can target specific channels per topic, which in practice targets specific teams.

What if the visitor asks for a human when nobody is online?

The offline experience triggers: the visitor sees your offline message and a form to leave contact details, which lands in your Inbox and email. Configure it under What Happens When Everyone Is Offline.

Does handoff work for WhatsApp conversations?

Yes. WhatsApp chats routed to the widget use the same AI and handoff settings, and escalated conversations land in Teams, Slack, Google Chat or the inbox. See How to Connect WhatsApp to ChatGPT.

Related Articles