Write chat routing rules on topic, sentiment, intent, VIP status, channel and business hours, and each escalation goes to the person or group who should take it, with the matched rule shown on the conversation.
A rule with at least one condition is checked on every new message while the AI is handling a chat. When it matches, the conversation moves to Waiting for agent and the AI does not reply to that message. At the Larkspur Hotel, an example 64-room hotel, a rule of Sentiment equals Angry hands a guest who is furious about a $28 parking charge to the Front desk group, so the complaint never gets a cheerful automated reply.
When the AI decides a person should take a chat, for example because the guest asked for a manager, it records why, and the same rules decide which agents are notified. The reason shows on the conversation as an AI escalated note, so Dana R. knows the story before she reads the thread.
Each routed conversation carries a Routing card with the rule that matched, who it went to and the signals at that moment. The rules list shows how often each rule has fired and when it last did, and marks a rule whose teammate or group has been deactivated as Target inactive.
The Messages tab holds three lines a customer can see during a handoff. Each one can be reworded with tokens such as {botName} and {companyName} ({agentName} on the takeover line) or switched off, and changes go live when you publish the bot. Here is how a handoff reads at the Larkspur.
A guest writes in, angry about a $40 late checkout fee. If a rule matches, the AI skips its reply and the guest gets that rule's own message, such as "I'm passing this to our front desk, who can see the charges on your room." If the AI raises the handoff itself, it works the Awaiting a human line into its reply instead.
The chat now shows Waiting for agent in the shared inbox. The AI stays on and answers anything else the guest asks, like whether breakfast comes with a King room or what parking costs, so nobody sits in silence while the front desk finishes checking someone in.
When Dana R. clicks Take over, or an admin assigns the chat to her, the guest gets the Agent takeover message with her name filled in from {agentName}. From then on the AI stays quiet and Dana writes the replies herself.
Once the fee is refunded, Dana clicks Release. The guest gets the Handed back to the bot message, and the AI handles the follow-ups, such as what time checkout is on Sunday. Switch any of the three Messages tab lines off if you would rather the change happen without a word.
What you can set on a rule under Bot > Handoff > Rules today, with the options the app offers.
Two things can start it. The AI hands over on its own when someone asks for a person, stays upset after it has tried to help, or needs something it cannot do or answer, usually offering a teammate first. Your rules can start it too: when a new message matches a rule with at least one condition, the conversation moves to Waiting for agent before the AI replies.
Book a demo with the topics your team always handles itself and who owns each one. We'll map them to handoff rules with you, show how the rule tester reads a demo conversation, and help word the messages your customers will see.
