Human Handoff in the Chat Widget
Human handoff lets an agent from your team take over a conversation that started with AI. The visitor keeps writing from the same chat window; they do not need to move to another app.
Creating the handoff request and the agent accepting the conversation are different stages. When the request is sent, the agent may not have joined the conversation yet.
Prepare Human Handoff
Open the Human Handoff section in the panel and enable the feature. Then review three areas:
- Agents: Adjust the team members who will take over conversations and their availability.
- Handoff Distribution: Determine how requests will be distributed, the working hours and the behavior applied when no suitable agent is found.
- Rules: Set the conditions that will start a handoff request, the messages and the notification preferences.
When Does Handoff Start?
Handoff is started according to the conditions you enable. The panel has options for the customer explicitly wanting to talk to a human, the AI being unable to answer, negative expressions, smart detection and custom keywords.
For example, the customer saying "Can I talk to an agent?" is an explicit agent request. If you use custom keywords, choose them carefully; broad-meaning words that occur frequently in everyday conversation can lead to unnecessary routing.
Test the rules with examples appropriate to your business's real conversations. A topic being complex or the customer making a negative sentence does not mean an agent should be reached in every case.
How Does an Agent Take Over the Conversation?
When a handoff request is created, it is offered to suitable agents according to the routing settings. Team members follow the requests from the panel; notifications are delivered according to the preferences you enabled.
When the agent accepts the request, they take over the conversation and reply from the Messages section. While the conversation is with the agent, the AI does not answer automatically. The visitor, meanwhile, continues the correspondence through the same widget.
Working Hours and Agent Availability
Adjust your working days, hours and time zone from the Human Handoff section. Set the hours you will provide live support here; also keep the working hours written in the knowledge base consistent with these settings.
An agent who is online, has handoff acceptance on, is not away and has available capacity can receive requests outside working hours too. If you do not want to receive handoffs, update the agent's availability or handoff acceptance status.
The AI assistant being able to answer outside working hours does not mean an agent is also ready at that time. Give the visitor information about live support that matches your team's real working arrangement.
If No Suitable Agent Is Found
Choose the behavior applied when no suitable agent is found from the routing settings. Adjust the wait, callback and leave-information options so your team can follow them.
If there is no suitable agent, the situation is explained to the visitor and the handoff request waits. When the visitor keeps writing, the AI can offer help. If contact information is missing for a callback, the visitor is asked for a phone or email.
Getting contact information from a visitor does not mean the callback happened. Your team needs to check these requests regularly and reach the customer.
If the Request Is Not Accepted in Time
In the Rules section, you can determine after how long the AI continues the conversation when the handoff request is not taken by an agent. You can also adjust the messages shown in the busy or timeout situation.
Let your message clearly tell the customer the next step. For example, state that they can continue getting help from the AI or leave contact information for your team to get back to them. Do not write a return time your team cannot meet.
Try Before Going Live
With an agent
Without an agent
Timeout
Off-hours
Leaving information
Detailed management
Related Topics
Human Handoff (Panel)β
Detailed management
Messagesβ
Following conversations
Customizationβ
Widget appearance
Was this page helpful?