AI chatbot human handoff, without starting over.
AI chatbot human handoff is the moment an AI agent stops answering and a person takes the conversation, ideally in the same thread, with the reason, the urgency and everything the customer already said. Cloodot works that way on WhatsApp, Instagram, Messenger, webchat and email: the agent decides when, rates how urgent it is, and routes the conversation to the team that fits.
No credit card required · 7-day free trial
I was charged twice for the same order. I want to talk to a manager.
Customer was charged twice for order #4471 and wants the duplicate refunded today. Both charges confirmed on the order; the refund needs a person.
- Priority
- High, money at stake
- Routed to
- Billing team inbox
- Assigned
- Megan Whitaker
Where AI chatbot handoffs usually go wrong
The classic failure is the loop: the customer types "agent" three times and the bot offers the same menu. The second is the cold transfer, where a person finally appears and opens with "How can I help you today?", and the customer explains everything again. The third is the handoff into nowhere, escalated at 11 p.m. into a queue nobody watches until Monday, with no word to the customer.
There's a quieter one too: two voices in one thread. If the bot keeps replying after a person joins, the customer gets contradictory answers. A handoff has to change who owns the conversation, not just add someone to it.
Getting it right comes down to three decisions: when the AI should stop, who should get the conversation, and what that person needs to know when they open it.
When should a chatbot hand off, and what should it pass along?
When: hand off when the customer asks for a person, when they're frustrated or writing in again about the same problem, when money beyond your policy is at stake, and for anything touching safety, legal questions, security or fraud. Don't hand off just because the agent lacks an answer to a routine question. That's a gap to fill in the knowledge base, not a job for a person.
In Cloodot you write the when in plain language as handover conditions on the persona, such as "customer disputes a charge" or "customer asks to cancel a contract". They sit alongside built-in triggers for an explicit request for a human, frustration and sensitive issues. If the agent gets stuck mid-reply, it sends your fallback message rather than leaving the customer hanging.
Who: at handover the agent takes itself off the conversation, so it stops replying, and moves it into the team inbox whose description fits the reason, such as Billing, Returns or Spanish-language support. If no team clearly fits, it goes to Queued. Auto-assignment can then give it to a specific teammate, by round robin or, in Auto (AI) mode, by matching the customer's message to each teammate's expertise notes, with optional filters that skip anyone away, offline or at their cap.
What: the teammate opens the same thread, not a ticket. The handover entry reads "Handed over by AI agent:" with a one-line reason and a priority from No priority to Urgent. Labels and custom fields the AI filled during the chat, an order number or an issue type, are in the sidebar, along with the contact's stage and summaries of their earlier sessions. If the customer wrote in another language, any message can be translated in place.
What Cloodot does here
Handover conditions
Plain-language rules on the persona decide the moments that go to a person, on top of built-in triggers for explicit requests, frustration and sensitive issues.
Explore AI agentPriority on every handover
The agent rates urgency on a six-level scale and records why. Your team can sort the inbox by it, so a disputed charge sits above a routine question.
Team inboxes and AI routing
On the Starter plan and above, handovers land in the inbox-enabled team whose description fits the reason, or in Queued when none does.
Explore Unified inboxAuto-assignment
Manual, Round robin or Auto (AI), with filters for availability, who has Cloodot open and a cap on open conversations per teammate.
Team Handovers insights
See which team finishes the work handed to it, which passes it along, where handovers came from, and how long pickup and resolution took.
Explore Analytics
Pitfalls to design out
After-hours handoffs need an honest message. Set your hours in Settings → Organization → Business Hours. The agent knows when your team is online and can tell a customer at 11 p.m. that a person will pick this up in the morning, instead of implying someone is typing.
Vague team descriptions route badly. "Support" tells the AI nothing; "Refunds, chargebacks and duplicate charges" does. The same goes for handover conditions.
Too many handoffs usually means thin knowledge, not a timid AI. Check the FAQ Unanswered tab and the handover numbers in Insights before loosening the rules. And if you want a hard rule, such as a key account always going to a person, build it as a workflow: an Assign to User step clears the AI the moment it runs.
Finally, give the conversation back when the hard part is done. Choose Handover to your AI agent from the Agent row and it carries on in the same thread.
Set up AI chatbot human handoff in Cloodot
5 steps
Write handover conditions
In Intelligence → Agent, open the Persona tab and add conditions under Handover to humans. Describe each moment the way you'd brief a new hire.
Describe your teams
In Settings → Organization → Teams, turn on Team inbox for the teams that take escalations and write a description of what each one handles.
Choose how teammates are picked
In Settings → Organization → Auto-assignment, pick Round robin or Auto (AI), set the filters, and keep Also assign when your agent hands over switched on.
Set business hours and a fallback
Add your hours under Business Hours, and write a fallback message in the persona's Advanced Settings that offers a person.
Test the hard cases
In the playground, act out an angry customer, a refund over the limit and a request for a manager, then Save & Deploy.
Questions teams ask
How does an AI chatbot know when to hand off to a human?
In Cloodot, from two sources. Built-in triggers cover an explicit request for a person, frustration, sensitive issues and similar escalations. Your own handover conditions, written in plain language on the persona, cover the rest, such as refunds above a limit or contract cancellations. When any of them matches, the agent hands over and records the reason.
Does the customer have to repeat themselves after the handoff?
No. The teammate works in the same conversation, with the whole thread above them, the agent's one-line handover reason, a priority, and the labels and custom fields the AI filled in along the way. Summaries of the customer's earlier sessions are in the sidebar too.
What happens if nobody is online when the AI hands off?
The conversation waits in the team inbox or Queued; it isn't given to someone your rules exclude. With auto-assignment on, when a teammate sets their status back to Available, a few of the oldest waiting conversations are routed to whoever qualifies. With business hours set, the agent can tell the customer when to expect a person.
Does the AI keep replying after a person takes over?
No. An AI handover clears the AI from the conversation, and so does assigning it to a teammate yourself or with a workflow's Assign to User step. From then on only your team replies until someone hands it back to the agent.
Can the human hand the conversation back to the AI?
Yes. In the conversation's Agent row, choose Handover to followed by your AI agent's name. The agent picks up in the same thread with the history intact. Snoozing a conversation the AI owns pauses it without unassigning it, and it resumes when the snooze ends.
Which channels does handoff work on?
WhatsApp, Instagram, Facebook Messenger, webchat and email, all in one inbox. Meta's 24-hour window still applies on WhatsApp, Instagram and Messenger, so a teammate replying more than a day after the customer's last message needs a WhatsApp template, or has to wait for the customer to write again on Instagram and Messenger.
Bring us your Monday.
Start free. Train a Cloodot agent on your own docs, your own tone, your own tools, and you'll walk away with a working deployment in your inbox the same afternoon.