Human Handover for a Website Chatbot: How to Set It Up So Customers Never Get Stuck

Here is a small bug we found in our own chatbot, and it teaches the whole lesson of this article.

A visitor typed: "I need to confirm something." The bot replied: "Hello! I'm Support. How can I help you today?"

It was not a joke. The bot looked for greeting words, and the word "something" contains "hi". So it decided the visitor was saying hello, and it answered with a greeting. The visitor had asked a real question. The bot made it look like nobody was listening.

The bug is fixed. But think about what the visitor saw: a bot that did not understand, and no way to reach a person. Most visitors would not try twice. They would close the chat, and the sale or the support request would be gone.

Every chatbot will get things wrong sometimes. What matters is what happens next. That is what "human handover" means: the moment a bot gives a conversation to a real person, and how well it does it.

What handover is, in one sentence

Handover is when the bot stops answering and a person takes over the same conversation, so the visitor does not have to start again.

The last part matters most. "Contact us on this email" is not handover. It is the bot giving up. Real handover keeps the chat, the history and the visitor's details together.

Why small businesses need it more than big ones

A big company has a large knowledge base and many agents. A small business has a bot that knows the basics and one or two people who know everything else.

That is fine, as long as the bot knows its limits. Think about the questions your customers really ask:

• "Is this item in stock in Abuja?"
• "I paid but I have not received my order."
• "Can you do it for less if I buy ten?"
• "Your page says one thing and my receipt says another."

A bot can answer the first kind with a good knowledge base. It cannot, and should not, decide a discount or investigate a missing payment. Those need a person. If your bot tries to answer them anyway, it will be confidently wrong, which is worse than being silent.

The four moments when a bot should hand over

Decide these in advance, and write them down.

1. The bot has no answer. This is the most common one. If the knowledge base does not contain the answer, the bot should say so and offer a person. It should not guess.

2. The visitor asks for a person. Words like "agent", "human", "speak to someone" are a clear request. Make the option visible in the chat window, so people do not have to find the magic word.

3. The topic is about money or trouble. Payments, refunds, complaints and anything where the visitor sounds upset should reach a person quickly. A bot saying "I understand your frustration" does not calm anyone.

4. The bot is going in circles. If the visitor repeats the same question twice, the bot is failing. Offer a person on the second try, not the fifth.

What a good handover looks like to the visitor

Here is a made-up conversation, to show the shape. It is not a real customer.

Visitor: I paid for my order yesterday but nothing has arrived. Bot: I'm sorry about that. I can't check payments myself, so I'll pass this to our team. Please add your name and email, and we will reply here and by email. Visitor: (adds name and email) Bot: Thanks, Ada. Your message has gone to our team. We usually reply within one working hour. (a few minutes later) Sam (Support): Hello Ada, this is Sam. I can see your order. Let me check the payment now.

Look at what the bot did right:

• It said plainly what it could not do.
• It asked only for what the team needs: a name and a way to reply.
• It gave a time expectation.
• When the person arrived, they introduced themselves by name and showed they had read the history.

Look at what it did not do: it did not ask Ada to explain again.

What a good handover looks like to your team

The visitor's side is half of it. The other half is how your team receives the chat.

They should see the whole conversation. An agent who joins blind will ask the same questions the bot just asked. Make sure the history comes with the chat.

They should know how to reach the visitor. If the visitor left an email, replies should go there as well as to the chat window, in case they have closed the page.

They should know who has it. This is the one most small teams get wrong. If three people can all see a new chat, two of them may answer at once and the visitor gets two replies that disagree.

The one-agent rule

The simplest fix for the "two people answer at once" problem is also the best one: a conversation belongs to one agent at a time.

Here is how it works in practice:

1. A new chat arrives. Everyone on the team is told.
2. The first person to answer, or to press "take", becomes its owner.
3. Everyone else can still read it, but they cannot reply.
4. If the owner cannot finish, they transfer it to another agent or to a department. Then the new person owns it.
5. When it is solved, the owner closes it.

This sounds strict, but it prevents a lot of embarrassing mistakes. It also tells the team at a glance which chats are being handled and which are waiting.

Two details make it work well:

• Show the agent's name to the visitor. "Sam is helping you" feels better than "Support team".
• Let the business owner step in. If an agent falls ill or goes offline with a chat open, the owner needs a way to take it over, or that visitor waits forever.

Departments, if you have more than a few people

Once you have more than three or four agents, group them. For example: Sales, Billing, Technical.

Then a transfer can go to a department instead of one person. The chat joins that team's queue, and anyone in the team can take it. This matters because the person you want may be away, and you do not want the chat stuck on their name.

Set a promise, and keep it

Visitors forgive a wait when they know how long it will be. They do not forgive silence.

• Pick a number you can actually meet, such as "within one working hour", and show it in the bot's handover message.
• If you are closed, say so and give the time you open. A bot that says "our team replies from 9am" is honest and kind.
• Make sure someone is looking at the chats list during those hours. A handover that nobody sees is worse than no handover.

How to set it up, step by step

Here is the order that works, whichever tool you use.

1. List your triggers. Write the four moments above in your own words and add any that are special to your business.
2. Write the handover message. One short message that says what the bot cannot do, what happens next and when to expect a reply.
3. Add your team. Names, email addresses and, if you use them, departments. Keep the list short and current. When someone leaves, remove them the same day.
4. Decide the visitor details to ask. Name and email is usually enough. Do not ask for more than you need.
5. Test the full path. Open the chat as a visitor and ask something the bot cannot answer. Check that the right people are told, that the first agent can take it, and that another agent cannot reply.
6. Test the exit. Close the conversation as an agent and send another message as the visitor. The bot should be back in charge.
7. Read the chats every week. The questions that reach a person are exactly the ones to add to the bot's knowledge base. Over time, the bot gets better and fewer chats need a person.

Five handover messages you can adapt

These are made-up examples. Change the words to match your voice.

• No answer in the knowledge base. "I don't have that information, and I would rather not guess. I'll pass your question to our team. Please leave your name and email so we can reply."
• The visitor asks for a person. "Of course. I'll connect you with our team now. What is your name?"
• A payment problem. "I'm sorry about this. I can't see payments from here, so I'll send it to our billing team. They usually reply within one working hour."
• Outside working hours. "Our team is offline now and back at 9am. Leave your message and it will be the first thing they read."
• After the agent closes the chat. "This conversation is closed. If you need anything else, just write here and I'll help or bring in the team."

Notice that none of them blames the visitor or pretends the bot can do more than it can.

How to know whether your handover works

You do not need a dashboard full of numbers. Watch three.

1. How many chats reach a person. If it is almost none, your bot may be hiding the option. If it is almost all, your bot knows too little and needs a better knowledge base.
2. How long visitors wait for the first reply. Look at the slowest ones, not the average. One chat that waited a day tells you more than ten that waited five minutes.
3. How many chats are still open after a week. These are chats nobody finished. Close them, or find out why they stalled.

Review these every two weeks. Each time, pick the three most common questions that reached a person and teach the bot to answer them.

Handover outside the chat window

Not every visitor stays on your page. Plan for these cases:

• They close the page. If you have their email, send the reply there as well, so they do not have to come back to see it.
• They return later. The chat should still be there with the history, so the next message continues where it stopped.
• They would rather use WhatsApp. Offer it as an option, and make sure the agent has the conversation so far.

Mistakes to avoid

• Hiding the human option. If it takes three clicks to find, people leave before they find it.
• Letting the bot argue. If a visitor has asked for a person, hand over. Do not ask them to try the bot again.
• No owner for each chat. If everyone can answer, nobody does, or everybody does.
• Handing over with no context. Sending an agent a blank chat wastes everyone's time.
• Never closing chats. Open chats pile up and your waiting list becomes noise.
• Treating handover as a failure. It is the feature that makes the bot safe to use. A bot that hands over well is a good bot.

How this works on XpiriaTech

The XpiriaTech chat widget includes handover. When the bot has nothing to answer with, the visitor is offered the team, with their question already filled in. Your support team gets an email with a private link, the first agent to answer takes the conversation, and only that agent can reply until they transfer it to another agent or to a department. Visitors see the agent's name, and the owner can step in at any time. Every conversation, including the ones the bot handled alone, appears in an inbox on the bot's page.

You can read about the widget on the AI chat page. Whichever tool you use, test the full path with the steps above.

Common questions

Does handover mean I need a full support team? No. One person is enough. The point is that the visitor reaches a human instead of a dead end. With one person, set honest hours.

Should the bot answer outside working hours? Yes, for what it knows. For everything else, it should take a message and say when you will reply.

What if nobody answers a handed-over chat? That is the worst outcome, so watch for it. Check the waiting list at the start and the end of each working day, and set the promise you can keep.

Can a bot hand over to WhatsApp? Many businesses do this: the bot collects the details, then continues on WhatsApp. If you do this, tell the visitor clearly, and make sure the conversation history is not lost.

The short version

Decide when the bot must step back, make the human option easy to find, give each chat one owner, show the agent's name, and keep your promise on reply time. A chatbot is not there to replace your team. It is there to handle the easy questions so your team has time for the ones that matter.