A WhatsApp AI agent should hand a conversation to a person the moment the customer asks for one, the request becomes risky, or the agent stops making progress. In WACRM the agent then switches itself off on that chat, leaves a private note for your team, and a teammate continues in the same WhatsApp thread from the shared inbox.
This guide is for anyone running, or about to run, a WACRM AI agent on their WhatsApp number: an online shop answering order questions, a clinic taking bookings, a training centre answering course questions. It explains why the handoff matters, exactly when it fires, and the three things a person does with it: pick up, take over and hand back.
Why does a WhatsApp AI agent need a human handoff?
Because WhatsApp requires one, and because customers notice when they are stuck with a bot. WhatsApp’s Business Messaging Policy (last updated 23 September 2026, checked 13 November 2026) allows automation when responding inside the 24-hour window, but says the business “must also have available prompt, clear, and direct escalation paths”. In-chat transfer to a human agent is one of the options it lists, next to a phone number, email, web support, a store visit or a support form.
What the policy asks of any automated WhatsApp reply
Automation is fine inside the 24-hour window, as long as the customer can reach a person promptly, clearly and directly. An in-chat handoff to your team inbox is the most direct of the listed paths, because the customer does not have to switch channel or repeat themselves.
AI agents that serve a business’s own customers are welcome on the platform. When WhatsApp changed its terms to bar general-purpose chatbots from 15 January 2026, Meta told TechCrunch the move “doesn’t affect businesses that are using AI to serve customers on WhatsApp”. A customer-service agent with a working path to people is exactly that kind of use.
There is also a plain business reason. An AI agent is very good at the repeat questions: opening hours, sizes, delivery times, the status of an order from your WooCommerce or Shopify store. It is the wrong tool for an angry customer, a refund dispute or a medical question. The handoff is what lets you automate the first group without gambling on the second. For the shop side of that first group, the WhatsApp AI agent for ecommerce playbook shows how the agent answers price, stock and order questions from your catalog.
When does the WACRM AI agent hand a chat to a person?
The WACRM AI agent hands over in four situations, and checks them before it sends anything. The table sums them up; the sections below explain each one.
| Situation | What triggers it | AI credit used |
|---|---|---|
| The customer asks for a person | A phrase such as “talk to a human” or “real person”, matched as whole words | None, checked before drafting |
| A high-risk action | The agent wants to do something that needs a teammate’s approval at high risk | The draft, which is kept for review |
| The agent repeats itself | Three proposed sends to one chat within five minutes | None, the agent stops first |
| The AI wallet is empty | The wallet cannot cover the next draft | None |
Source: WACRM product behaviour, checked 13 November 2026.
When the customer asks for a person
If a customer writes “I want to talk to a human”, “agent please” or the same request in another language, the agent stops before it drafts a reply. WACRM recognises these requests in English, Vietnamese, Spanish, Portuguese, Indonesian, Hindi and Arabic, and matches whole words and phrases only, so the same letters inside a longer word do not trip it by accident. Because the check runs before any AI call, the handoff costs no AI credit.
When the agent wants to do something risky
Some actions always need a teammate’s approval at high risk: sensitive tools, and a free-text message outside the 24-hour customer service window, which WhatsApp does not allow without a template. When the agent reaches one of those, it hands the chat over and leaves its proposal in the approval queue, so a person can approve, edit or reject it with the full context.
When the agent repeats itself
An agent that proposes three sends to the same chat within five minutes stops itself. That loop guard, also listed on the AI agents page, now hands the chat to a person instead of just going quiet, so a confused customer is not left waiting.
When the AI wallet runs out
AI usage is paid from your WACRM wallet and never goes negative. If the wallet cannot cover a draft, every conversation the AI was handling in that project moves to the team queue with the AI switched off. Top up, then hand the chats you want back to the AI.
What happens in the inbox at the moment of handoff?
Four things happen together, in one step, so nobody sees a half-finished state: the AI switches off on that conversation, the chat goes to a person, a private note is written, and the team is notified.
1 · Trigger
Customer: “Can I talk to a real person about my refund?“
2 · AI off on this chat
The agent stops before drafting. Nothing is sent to the customer.
3 · Note and system line
Private note: reason, unsent AI draft, last six messages.
4 · Teammate notified
Push to the teammate on the chat, or to the whole team.
Who gets the chat. If a teammate was already on the conversation, it stays with them. Otherwise it goes to the team queue in the team inbox, where anyone with access to that chat can pick it up.
The private note. The note opens with the reason, for example that the customer asked for a person. If the agent had drafted a reply, the draft is included and marked as not sent. Then come the last six messages from the customer, your team and the AI, each cut to 200 characters, so the teammate can answer without scrolling. Notes are internal and never go out on WhatsApp.
The system line. The thread shows a line saying that the AI handed the conversation over, to whom and why, so the history stays readable weeks later.
The notification. A push notification goes to the teammate who now has the chat, or to every teammate with push notifications turned on when it went to the team queue. The AI stays off on that chat until a person hands it back.
How do you take over a chat from the AI yourself?
You do not have to wait for a trigger. Any teammate who can see a conversation the AI is handling can take it over in one action, for example when they spot a VIP customer or a complaint brewing.
AI agent is handling this conversation
Take over TThe jacket arrived in the wrong size. I need it before Friday.
Sorry about that! Which size would you like instead?
- Open the conversation in the WACRM inbox. While the AI holds it, a banner at the top reads that the AI agent is handling this conversation.
- Click Take over, or press T on your keyboard. The shortcut is ignored while you are typing in the reply box, so it never fires by accident.
- Check the header: the chat now shows you as the person on it, and the AI mode is off.
- Reply to the customer. The thread records a system line that you took over the conversation.
If the chat belongs to another teammate, your role decides whether you can take it; the team members guide explains roles. Replying yourself without pressing Take over also pauses the AI on that chat: the composer then says “AI paused here until” a time, or until someone hands it back, so you always know whether the agent can come back on its own.
One thing to keep in mind: a reply the agent proposed before you took over stays in the approval queue. It only goes out if a person approves it, so clear the queue for that chat if the situation has changed.
How do you hand the conversation back to the AI?
When the human part is done, switch the conversation’s AI mode back on with Hand back to AI. The handoff reason is cleared, and the agent answers the next customer message as usual.
Good moments to hand back:
- The refund or complaint is settled, and the customer’s next questions are routine ones.
- You confirmed an exception, such as a late delivery date, and wrote it in a note for the agent’s context.
- The AI wallet has been topped up after an out-of-credit handoff.
Leave the AI off for conversations that should stay personal: a key account, a sensitive health topic at a clinic, or a customer who asked not to talk to a bot. The workspace-wide AI pause switch is the other tool: it stops every AI action in the project at once, which is useful during a stock problem or a price change.
What should the handoff note tell your teammate?
WACRM writes the note for you, but the reply quality still depends on what your team checks before writing. Use the note plus this short checklist so the customer never has to repeat themselves:
Before you reply after a handoff
- Read the reason line first: a human request, a risky action, a loop or an empty wallet each need a different first reply.
- Check the unsent AI draft, if there is one: often half of it is right and saves you typing.
- Look at the contact’s linked orders or bookings before quoting a status or a date.
- Check whether the store has already acted: an order marked Refunded or On hold changes what you should promise.
- Open with a line that a person is now replying, then answer the actual question.
- Add your own note when you decide something the AI should know if you hand back.
Store status matters most in refund and payment handoffs. WooCommerce’s order status documentation (checked 13 November 2026) defines Processing as paid and awaiting fulfilment, On hold as awaiting payment confirmation, and Refunded as fully refunded by an admin or shop manager, so read the status before you promise anything.
The customer’s view matters here. They never see the handoff note or the system line, only your next message. A short “Hi, this is Mai from the team, I’ll sort out the size swap for you” is enough to show that a person has the conversation now. AI assist can still suggest that reply, and your teammate reads, edits and sends it.
How does the 24-hour window affect a late human reply?
It decides what you can send. Meta’s pricing documentation (checked 13 November 2026) says a 24-hour customer service window starts when the customer messages you and resets each time they message again. Inside it you can send free-text messages; once it closes, only approved template messages can reach the customer.
| When your teammate replies | What they can send | What to do |
|---|---|---|
| Within 24 hours of the customer’s last message | Any free-text reply | Answer directly in the thread |
| More than 24 hours after it | Approved template messages only | Send a template that reopens the conversation |
Source: Meta for Developers, Pricing on the WhatsApp Business Platform, checked 13 November 2026.
That is why a handoff should be picked up fast. A chat handed over on a Friday evening and opened on Monday needs a template first; the first template guide shows how to create and send one, and the free service messages guide explains what Meta charges for replies inside the window. The same rule is the reason a free-text AI reply outside the window always waits for a teammate’s approval.
How do you set up handoff before going live?
Handoff is on with every WACRM AI agent; what you set up is the team around it. A short checklist before you let the agent answer customers:
- Invite the teammates who will pick up handed-over chats, with roles that let them see the conversations.
- Ask each of them to turn on push notifications, so a handoff to the team queue reaches someone quickly.
- Leave AI auto-send off for the first days, so every proposed reply waits for approval while you watch where the agent hands over.
- Agree who watches the team queue at each hour your customers write, including evenings and weekends.
- Prepare one approved template for replies that miss the 24-hour window.
- Top up the AI wallet so the agent does not hand everything over for lack of credit.
If you would rather have this configured for you, Quick Setup connects your number, gets 5 starter templates approved and trains your team in a 30-minute call, and pricing lists what each plan includes. For more how-to articles across WhatsApp, browse all WACRM guides.
Ready to let an AI agent take the first reply while your team keeps the last word? See how WACRM AI agents answer, book and hand over on your WhatsApp number.
