How to get a chatbot human handoff right before the customer gives up

A good chatbot human handoff keeps a customer you’d otherwise lose to a stuck bot. In a Gartner survey of 3,566 customers, 87% said companies using AI for service must give people a route to a real person. Here’s when to hand over and how to do it.

A handoff is the moment the bot passes the conversation to a person. Get it wrong and the customer pays first, then you do. After a bad service experience, 34.9% of US adults in Parloa’s 2026 Consumer Patience Index moved to another brand and 27% aired their complaint in public, through reviews or social posts.

When should a chatbot hand a chat to a person?

Hand the chat over as soon as the bot can’t answer with confidence, the customer asks for a person, or the conversation turns to money or account security. Sinch’s guidance on AI agent escalation treats money and security as human work however much the bot happens to know, and that’s a sound default for a small team too.

Don’t wait long. In Parloa’s survey of 1,001 US adults, 55.5% said they’ll stick with a bot for under three minutes before they want a person, and only 5.5% would stay with it past 10 minutes.

Count failed answers, not minutes. The same Parloa survey found that one in ten people quit as soon as a bot makes them say something again, and another 60.1% will put up with two repeats at most. Set the bot to hand over after its second miss, not its fifth.

Watch the tone, too. Sinch points to clipped replies, typing in capitals and skipping the please and thank you as early warnings, and advises passing the chat on before the customer has to ask. Once a customer starts typing in capitals, the bot’s turn is over.

Some topics should skip the bot entirely. Billing disputes top the list in SurveyMonkey’s research on AI customer service, where 85% of Americans would rather deal with a person. Data security follows at 78%, then fixing a faulty product at 76% and questions about price at 71%.

Eric Keller, a Gartner analyst who covers customer service as a senior director, puts it plainly: “Service leaders should not use GenAI as a mandatory first step for every issue.” Give the customer a visible way out from the first message, not only after the bot has failed.

What should the bot say at the handover?

The bot should say plainly that it’s passing the chat to a person and what happens next, so the customer knows the wait has an end. A vague holding message reads like one more dead end.

Honesty here is about trust. SurveyMonkey found that 23% of Americans stop trusting a business once a bot keeps them going in circles, and 11% do over canned, generic replies. Its advice is for bots to say what they are and be upfront about the problems they can fix.

A good handover message does three jobs:

• It confirms what the customer asked, in a few words, so they know they were understood.
• It says a person is taking over, and whether that person will reply in the chat or by email.
• It gives a realistic wait, or says honestly that the team will reply as soon as someone is free.

Write a different message for each reason the bot hands over. A customer who asked for a person wants a quick yes. A customer the bot couldn’t help wants to hear that the question is going to someone who can answer it. LenoChat keeps those two messages apart and lets you rewrite both, as the guide to what happens when a visitor asks for a human agent shows.

What does the person taking over need to see?

The person taking over needs the whole conversation, including the message that triggered the handoff, so their first reply picks up where the bot stopped. Sinch makes a similar case: an agent who can’t read what came before slows the customer down rather than helping.

Repeating yourself is one of the most disliked parts of support. US adults in Parloa’s survey ranked a bot that doesn’t understand them as their top complaint, at 25.9%, and having to give the same details twice at 12.2%. A handover that asks for the order number again hits both.

The gap between what customers want and what they get is wide. SurveyMonkey, citing research published in 2025, reports that 78% of consumers care about being able to move from the bot to a person, yet only 15% get a smooth handoff.

Before the agent types a word, they should be able to see:

• The customer’s first question, in their own words.
• Every answer the bot gave, so they don’t repeat a wrong one.
• Why the bot handed over: no answer, a request for a person, or signs of frustration.
• Any details already collected, such as an order number or email address.
• The channel and the page the customer wrote from.

Then the first reply should prove the agent has read it. Open with the customer’s actual problem and your next step, not a fresh greeting that asks how you can help.

What should happen when nobody is online?

When nobody is online, the bot should say so and take a message with a reply time, instead of placing the customer in a queue no one is watching. A queue is the list of chats waiting for an agent, and overnight it can sit untouched until morning.

Nights and weekends are where a bot earns its keep. In SurveyMonkey’s December 2025 study, just 8% of Americans chose AI agents over people, and 41% of that small group said it was because the bot is always there. After hours, the bot’s job is to be there and be honest, not to pretend a person is on the way.

A live transfer works better than a contact form during staffed hours, because a person can reply while the customer is still on the page. After hours, a contact form beats a transfer, because the customer leaves knowing when they’ll hear back. A live transfer is best for business hours. A contact form is best for nights, weekends and holidays.

If your chat tool lets you set separate handoff rules for business hours and after hours, use that split. Keep the form short: name, email and the question. Then promise a reply time you can actually meet, such as first thing the next business day.

How do you know your handoffs are working?

Your handoffs are working when the questions that trigger them shrink week by week and the chats that reach a person end with an answer. Don’t judge the bot by its containment rate alone, the share of chats it finishes without a person. Sinch warns that containment and deflection figures score a bot higher for blocking access to your team.

Once a week, read the chats the bot handed over. Group the questions, write a proper answer for the ones that repeat, and add them to what the bot knows, the same way you did when setting up an AI agent chatbot. Refunds, billing disputes and account access stay with a person, as covered in deciding what to automate and what to keep human.

Do this well and customers stop fighting the bot. In Parloa’s survey, 53.6% confessed they deliberately try to dodge chatbots, yet 85% said they’d likely use one that fixes their problem nine times out of ten. The difference is a bot that knows when to step aside.

Sabrina Leblanc, who heads sales and customer success at SurveyMonkey as a senior vice president, puts the balance this way: “To win, teams must keep humans in the loop – using AI to accelerate the work, not replace the relationship.” Start with the handover rules and the two handover messages this week. They take an afternoon, and they decide whether a stuck customer stays or leaves.