AI-to-human handover on WhatsApp: when the agent should stop, what the customer should see, and how you take over
The moment that decides whether customers trust your AI agent is not the first answer. It is what happens when the agent cannot help. A clean handover feels like being passed to a colleague; a bad one feels like shouting at a machine. Here is the handover design that works, and how to set it up so it happens without anyone watching.
Quick answer
Hand over when the customer asks for a person, when the conversation turns to complaints, money, medical or legal matters, or when the agent is not sure. The customer should be told plainly that a person will reply here. The owner gets an alert, replies in the same chat, and the agent stays silent on that conversation until the human goes quiet for a period you choose. ReplyKit does all of this by default.
The five triggers
- The customer asks. "Can I talk to someone?" ends the automation on that chat, no exceptions and no "are you sure?".
- The topic is off-limits. Complaints, refunds, anything medical, legal or financial, anything about an order or case already in progress. You name these in the instructions.
- The agent is unsure. No relevant knowledge for the question. Better to hand over than to guess. ReplyKit also logs the question so you can teach it.
- Loop guard. A burst of replies in one chat trips the guard, and the chat goes to a person rather than looping.
- Provider failure. If the AI provider is down, the customer gets a holding message and the chat goes to the inbox, rather than silence.
What the customer should see
Three things make it work. It is honest: no pretending a person is already there. It is specific about where: here, in this chat, not a phone call or an email. And it sets an expectation without a promise: "shortly", not "within five minutes". Outside your hours, add the next step: "We open at 9am and will reply first thing." Never leave a handover with silence.
How the human takes over
The chat appears in the inbox flagged as waiting, and the owner gets an email with a link straight to it. Anyone on the team replies in the same conversation; the customer sees a normal WhatsApp message from the business number. From the first human reply the agent is silent on that chat, so there is never a bot and a person talking at once. The inbox shows who is speaking, and a filter lists everything waiting for a person. The customer never has to repeat themselves, because the human sees the full transcript, the lead's card and any photos.
Returning the chat to the AI
Once the human issue is resolved, the chat should go back to the agent so routine follow-ups are handled automatically. Two ways: click "hand back" in the inbox, or let it return on its own after the human has been quiet for a period you set in Settings, from one hour to 24, or never. Most businesses pick four hours. A customer who asks for a person again gets one again; the rules do not change because they were handed back.
What to watch
- Handover rate. Too high and the agent lacks knowledge or the boundaries are too tight. Under 15 percent is typical for a well-taught agent.
- Waiting time. How long handed-over chats sit before a reply. The weekly digest tells you how many are waiting.
- Reasons. The agent records why it handed over. If "unsure" dominates, teach it; if "complaint" dominates, that is a business signal, not an agent problem.
Frequently asked questions
When should an AI agent hand over to a human?
When the customer asks, when the topic is a complaint, money, medical or legal matter, when the agent is unsure, when the loop guard trips, or when the AI provider fails.
Can the AI and a person reply at the same time?
Not in ReplyKit. The first human reply pauses the agent on that conversation until it is handed back or the human goes quiet for the period you set.
How does the team know a chat is waiting?
The inbox flags it, an email goes to the owner with a link, and the weekly digest counts anything still waiting.
Does the customer have to repeat themselves?
No. The person sees the full transcript, the lead's details and any photos before replying.
Can I turn off automatic return to the AI?
Yes. Set it to never in Settings and hand chats back manually.
See a handover happen
Message the live demo and ask for a person. Then set your own up in 15 minutes. 7-day free trial.
Sources and further reading
- WhatsApp Business Messaging Policy · opt-in, opt-out and prohibited-use rules every business must follow
- Harvard Business Review, The Short Life of Online Sales Leads (Oldroyd, McElheran, Elkington, 2011) · firms that responded within an hour were about seven times more likely to qualify the lead
- ReplyKit Terms of Service, section 4 · your responsibility for AI-generated messages and human oversight
ReplyKit is an independent product and is not affiliated with or endorsed by Meta or WhatsApp. This article is general guidance for running a small business, not legal, financial or regulatory advice.