Guide · Guide
Chatbot to human handoff: how to pass a chat to your team without losing the customer
By Dev Patel, founder of AgeDash. Published . 7 minute read.
A good chatbot to human handoff has five parts: clear rules for when the AI hands over, one short check that the customer really wants a person, an alert that reaches someone on duty, a takeover where the person sees the whole chat, and a clean return to the AI when the person is done. The customer should never have to repeat themselves, and never be stuck talking to a bot that cannot help.
When should the chatbot hand over?
| Trigger | What the AI does |
|---|---|
| The customer asks for a person | Asks once what it is about, then hands over if they still want a person |
| A refund, complaint or legal question | Hands over straight away, because your rules say a person decides |
| The customer is upset | Apologises briefly and hands over, without arguing |
| The AI cannot find the answer twice | Says so and offers a person or a call back |
| A large order or a quote | Hands over, because a person closes these better |
Write these rules down with your team. They are the most important part of the setup, and they are easy to change later.
Ask once, then hand over
Many people type "human" out of habit, before they have even asked their question. One short question, such as "Sure, what is it about?", often solves it on the spot. But only once. If they ask for a person again, the AI should confirm and hand over, without trying to keep them. You can play this exact flow in the human handoff demo.
Alert the right people, the right way
A handoff is only as good as the alert. Decide three things:
- Who takes chats, and when each of them is on duty.
- How each person is told: a pop-up in the dashboard, a WhatsApp message, Slack or Teams, or a phone call with a recorded message for the person on duty.
- What happens when nobody takes the chat within a few minutes.
Send the alert to everyone on duty, and let the first person to take the chat have it. The others should see at once that it is taken, so two people never answer the same customer.
What the person needs to see
- The whole conversation so far, including what the AI already tried.
- Who the customer is, if known: name, order, booking or ticket.
- Why it was handed over: asked for a person, a refund, an upset customer.
- How long the customer has been waiting.
With this in front of them, the person can start with "Hi, I am Sana, I have read your chat" instead of "How can I help?". That one sentence is what customers remember.
The AI steps aside, then comes back
While a person is in the chat, the AI must stay quiet in that conversation. When the person ends their part, the AI can take over again for what is left: sending the ticket number, a reminder, or a short feedback question. It should know what the person did, so it does not offer the same help again.
After hours
A handoff needs someone to hand over to. Outside your hours, the AI should say plainly when a person will be back, take a phone number or email, and create a ticket or a call back. A customer who knows when they will hear from you waits calmly. One who waits in silence leaves.
Numbers worth watching
- Handoffs per day, and the reasons, so you can teach the AI the common ones.
- Time until a person takes the chat.
- Chats nobody took, and what happened to them.
- How many chats went back to the AI after a person was done.
The same idea on the phone
Voice agents hand over the same way: the AI answers the call, solves what it can, and passes the rest to a person live or as a call back, with the details on their screen. See the AI phone support agent.
Common questions
Does the AI keep replying while a person is in the chat?
It should not. A proper handoff pauses the AI for that conversation until the person ends their part, then lets it continue.
What if nobody on my team is free?
Set a time limit. After it, the AI tells the customer, takes their number or email, and creates a call back or a ticket, so nobody is left waiting.
Can my team reply from WhatsApp instead of a dashboard?
It can be set up that way, but a dashboard is better for more than one person, because everyone sees who has taken which chat.
Should the chatbot say it is an AI?
Yes. Customers accept an honest AI assistant much more easily when a person is clearly one message away.