A chatbot earns trust by knowing its limits. This guide to chatbot human handover covers the triggers that should bring in a person, the context an agent needs, queues and business hours, what to tell the customer and how to measure the results.
Key takeaways
Every chatbot reaches the edge of what it should handle. A refund dispute, a frustrated customer on their third attempt, a question about a prescription, a long-standing client who simply wants a person. What happens in that moment decides whether the customer remembers your bot as helpful or as a wall.
Chatbot human handover is the move from automation to a member of your team, with the conversation and its context intact. This guide covers when the bot should step aside, what the agent needs to receive, how queues and business hours shape the experience, what to tell the customer, and how to measure whether handovers are working.
A handover has three parts. The bot recognizes that a person is needed, the conversation moves to the right agent with everything gathered so far, and the customer knows what will happen next. When the agent finishes, the conversation returns to the bot or closes cleanly.
Two shortcuts undermine it. A cold transfer, where the customer is sent to a phone number or a contact form and has to start again, throws away the context the bot collected. A hidden handover, where it is unclear whether a bot or a person is replying, erodes trust the moment the customer notices. Say plainly when a person joins and when the bot is back.
Good escalation rules combine what the customer says, how the conversation is going and who the customer is. These are the triggers most service teams need:
| Trigger | Typical signals | What the bot does |
|---|---|---|
| Customer asks for a person | "Agent", "human", "speak to someone", or the same request in Arabic | Hands over at once, without one more article first |
| Low confidence | The best intent match falls below your threshold | Asks one clarifying question, then offers a person |
| Repeated failure | Two fallbacks in a row, or the same question rephrased | Stops trying and hands over with the failed attempts attached |
| Frustration | Negative sentiment, capital letters, repeated messages, words such as "useless" | Apologizes briefly and hands over |
| Complaint | "Complaint", a request for a manager, a threat to cancel | Routes to the complaints owner, not the general queue |
| Payments and refunds | Disputed charges, failed payments, refund amounts, bank details | Hands over; the bot may send a secure payment link but does not decide disputes |
| VIP or high value | A tag in your CRM (customer relationship management system), a contract tier or an order value above a limit you set | Routes to the account manager or a priority queue |
| Sensitive topics | Health, legal, account security, bereavement, anything your policy names | Mandatory handover with a carefully worded message |
Two rules make triggers work in practice. Explicit requests always win: a customer who asks for a person should never have to ask twice. And every trigger should appear in your reporting, so you can see which ones fire most and whether they fire for the right reasons.
A confidence threshold is not set once. Start with your platform's default, review the conversations answered with low confidence, and adjust. Too low, and wrong answers slip through; too high, and the bot hands over conversations it could have handled. A weekly look at the queue of unanswered and low-confidence questions turns repeat patterns into new content.
Nothing undoes a good bot faster than an agent who asks the customer to start again. The fix is a handover packet that arrives with the conversation, so the agent starts at the solution rather than at "How can I help?". It should include:
Where the bot connects to your ERP (enterprise resource planning system) or CRM, link the record rather than copying it, so the agent sees the live order or invoice. Our chatbot buying guide lists the questions to ask any vendor about what the agent actually receives.
A handover is only as good as the queue behind it. Route each conversation by topic, language and channel to the team that can resolve it, and decide in advance what happens when that team is busy or closed.
The handover message matters as much as the routing. It should say who is taking over, that they can see the conversation, when to expect a reply and whether the customer needs to stay on the channel. A realistic expectation that is met beats an optimistic one that is missed. During opening hours, something like this works:
I am passing you to our returns team. They can see this conversation and your order details, so you will not need to repeat anything. The team is online until 6pm and will reply here.
Out of hours, the message changes but the honesty does not:
Our team is offline now and back at 9am. I have saved your request with your order number and photos, and a colleague will message you here in the morning.
Write these messages natively in each language you support. When the agent joins, they should introduce themselves by name, and when the conversation returns to the bot, the bot should say so.
Here is how the pieces fit together for an online retailer that answers on WhatsApp.
Handovers are not a failure metric. Some conversations should reach a person, and a bot that never hands over is often trapping customers. The goal is the right handovers, quickly, with nothing lost on the way. Track these measures by trigger, channel and language:
Tip: Ask agents to tag every handed-over conversation with one outcome, such as "needed a person" or "bot could have answered". Reviewed weekly with the agents, those tags become a short, specific list of what the bot should learn next.
StarBot, our conversational AI chatbot, is designed around handover. Escalation triggers cover keywords, sentiment, repeated misunderstanding and customer requests, and you decide when handover is mandatory, for example for sensitive topics or high-value requests. Conversations are routed by topic, language, channel and business hours, and the agent receives the full transcript and captured details in your existing helpdesk or a shared team inbox. Out of hours, StarBot captures the request with a promised call-back, and the bot resumes automatically when the agent closes the conversation.
StarBot also connects to StarBiz and to Odoo through OdooStar, so the bot can look up the order, invoice or CRM record before it hands over, and what it finds travels with the conversation. We agree your escalation rules with you, test them against sample conversations and refine them during the pilot. Talk to us, call +971 55 973 4524 or email [email protected].
Star Bit Solutions is a software company in Ajman, UAE, founded in 2014. We implement and customize Odoo Enterprise and Odoo Community, and build StarBiz (AI-driven ERP), StarHub (life super app), StarBot (conversational AI), StarLearn (adaptive learning) and StarSense (AI and IoT platform). About us
Escalation triggers, routing by topic, language and hours, and the full transcript passed to your agents.
Learn MoreOrders, deliveries, invoices, stock and leads that StarBot can look up before it hands a conversation over.
Learn MoreStarBot for order inquiries, alongside StarBiz sales, inventory and finance.
Learn More
Language switching, Gulf dialect, Arabizi, right-to-left layouts, tone and bilingual handover: a practical design guide for Arabic and English chatbots.
· 10 min read
Ten criteria, the red flags to watch for and a pilot plan you can run in a few weeks.
· 8 min read
Containment, resolution, satisfaction, handover, first response, fallback and conversions defined, with vanity metrics to drop and a monthly review routine.
· 9 min read
Hand over whenever the customer asks for a person, and when the bot's confidence is low, it has failed twice in a row or the customer is clearly frustrated. Complaints, refund disputes, payment problems, sensitive topics such as health or legal questions, and VIP or high-value customers should also reach a person. Make explicit requests immediate, so nobody has to ask twice.
The reason for the handover, the full transcript in the original language, details already captured such as order number and dates, the customer's identity and whether it was verified, what the bot already tried, the channel and language, any attachments and priority flags such as VIP status. Link live ERP or CRM records rather than copying them into the chat.
Say so honestly, capture the details and promise a reply or call-back at a time your team can actually meet. Create the case so it is first in the queue when the team returns. On WhatsApp, free-form replies are only possible within 24 hours of the customer's last message, so keep an approved template ready for later follow-ups.
Not necessarily. Some conversations should reach a person, and a bot that never hands over is often trapping customers. Look at why handovers happen: frequent handovers for questions the bot should answer point to content gaps, while handovers for complaints, disputes and sensitive topics show the bot doing its job. Track the rate by trigger rather than as one number.
Track the handover rate by trigger, the time from handover to the first human reply, how many customers leave before an agent answers, how often agents ask for details the bot already captured, and resolution and satisfaction after handover. Review a sample of handed-over conversations every week with your agents to find unnecessary and missed handovers.
Tell us how your team handles complaints, refunds and VIP customers. We will design StarBot's escalation rules around them and test them with you before launch.