Conversation & Agent DesignGuide

When Your Chatbot Should Hand Off to a Human

In this article

There are four triggers. When one fires, your chatbot hands off to a human. Three triggers are about the message. They are money, mood, and law or safety. The fourth is the second time the bot misunderstands. Hand off on the first trigger. Send the whole chat with the handoff. Tell the customer that a person is on the way.

That is the rule. This post shows you how to set it up. Test the rule on a real message. Use our free Handoff Checker.

Why most chatbots hand off too late

Most bots try to keep the customer inside the bot. That is the wrong goal. Customers already know how long they will wait.

Five9 wrote its 2026 Consumer CX Report. It surveyed 3,000 people. They live in the U.S., U.K., and Germany. 92% give an AI agent under 10 minutes. Then they want a person. The top reason to quit the bot was repeated misunderstanding. 46% named it.

Customers do not hate AI. They hate being stuck. Verint surveyed 5,000 U.S. consumers in early 2026. 61% said they prefer a human agent over AI. But 69% of that group would use self-service. The condition: it fully solves their problem. The same report has one more number. After one terrible experience, 79% would leave for a competitor.

The math is simple. Some bots hold on too long. That costs you the customer.

What are the four handoff triggers?

A handoff is the moment a chatbot stops answering. The bot passes the customer to a person. The chat so far goes too. We use four triggers. If one trigger fires, the bot hands off. The bot does not try one more thing.

  1. Money. Refunds, disputed charges, and cancellations. Also anything that changes what a customer pays. A bot can explain a policy. A person decides the refund.
  2. Mood. Anger, fear, grief, or a plain “I want a person.” The words are usually clear. Examples: “this is ridiculous,” “I’ve asked three times.” Or: “someone used my card.”
  3. Law or safety. Legal threats, medical questions, harassment, or a possible data breach. A person must handle these. You often need a record too.
  4. The second miss. The bot misunderstood once. It tried again and missed again. Stop there. On a third try, customers quit.

Here is an example: “I need to cancel. I lost my job.” Two triggers fire at once: money and mood. Do not send a cancellation form. Send a person, fast. That person already knows what the customer said.

Four cards labeled Money, Mood, Law or safety, and The second miss, each with example customer phrases that should trigger a chatbot handoff to a human.
If any one of the four fires, the bot hands off. It doesn't try one more time.

You can try this example in the Handoff Checker. Type a message. Say how many times the bot missed. Say if a person is on shift. The tool shows which triggers fire. It also shows what the bot should say next.

What makes a chatbot handoff good?

The handoff is half the job. A bad handoff drops the customer into a queue. Then the customer must start over. Look at the Five9 survey. About a third of people had to repeat information. They had already given it to the bot. Verint calls this the handoff problem. Customers count that as a failure. This is true even when the person fixes the issue.

A good handoff does three things:

  1. The context goes with the customer. The person gets the full chat and the customer’s name. The person also gets the trigger that fired. The customer never types it twice.
  2. The bot says what happens next. “I’m connecting you to Dana on our team. I’ve sent her what you told me.” Do not say only “transferring.” Do not go silent.
  3. The customer can see a person from the start. Five9 also looked at trust in AI. With a visible human option, trust more than doubled. It went from 26% to 55%. Of those surveyed, 71% want to know when they talk to AI. They said it is very or extremely important. A hidden exit makes customers trust the bot less.
Two chat transcripts side by side. In the bad handoff, the bot says it is transferring and the human asks the customer to explain again. In the good handoff, the bot names the person, passes the context, and the human picks up where the bot left off.
Same customer, same problem. The second bot passed the context, so the customer only told the story once.

How do you set up handoff rules without a developer?

You do not need code for this. You need a one-page document. You also need one hour with your support lead. Do these four steps:

  1. Write the four triggers. Use one heading for each trigger. Under each heading, list real phrases from your own chats.
  2. Build a test set from 20 past chats. Pick the chats that went badly. Read each chat. Mark the line where a person should have stepped in. These marks show the words your customers really use.
  3. Decide where the handoff goes, by hour. During business hours, send it to a live person. After hours, give a callback. Name the person and set the time. Example: “Dana will call you before 10 a.m. tomorrow.” Some bots promise a person at 9 p.m. with nobody there. That is worse than no handoff.
  4. Measure two numbers every week. Count the chats that hit a trigger. Count the customers who repeated themselves after the handoff. The first number shows that the rules fire. The second number shows that the context arrives.

For the weekly check, use our rubric. It is in How to QA AI Writing Without Code. The same rubric works for a chatbot.

A one-page handoff rules card with three sections: the four triggers with example phrases, where the handoff goes by hour of day, and the two weekly numbers to track.
One page, written by the person who runs support. That's the whole system.

Where we land on this

At Intelligent CX Consulting, we judge a chatbot by one test. Did the customer get the right outcome? Time spent in the bot does not count. A handoff is the AI working as designed. The bots we build hand off on the first trigger. They send the context. They name the person who picks up.

For more depth, read our guide. It covers the seven moves a bot can make on any turn. It says when to ask or verify. It also says when to stop instead of handing off. Does your bot keep missing for no clear reason? Start with how fallback flows work. Want help to write your own one-page rules? Reach out.

FAQ

When should a chatbot hand off to a human? Hand off when a message is about money or mood. Also hand off for law or safety. Also hand off the second time the bot misunderstands. Hand off on the first trigger that fires. Do not wait for a third try. That is where customers quit.

How long will customers wait for a chatbot before they want a person? Not long. Five9 surveyed 3,000 consumers in 2026. Of them, 92% give an AI agent under 10 minutes. Then they want a person. The top reason to leave the bot was repeated misunderstanding. That reason got 46%.

Do customers actually prefer humans over AI? Most do, with one catch. Verint surveyed 5,000 U.S. consumers in 2026. Of them, 61% prefer a human agent. But 69% of that group would use self-service. The condition: it fully solves their problem. Customers want the problem solved.

What should happen during a chatbot handoff? Three things go to the person. These are the full chat and the customer’s name. The third is the trigger that fired. The bot tells the customer who is picking up. The customer never repeats the story.

Should my chatbot tell customers it’s AI? Yes. Five9 asked if people want to know when they talk to AI. Of them, 71% said it is very or extremely important. Customers could sometimes see a human option. Then trust in the bot more than doubled. Clear labels help the bot earn trust.

Frequently asked questions

When should a chatbot hand off to a human?

Hand off when a message is about money or mood. Also hand off for law or safety. Also hand off the second time the bot misunderstands. Hand off on the first trigger that fires. Do not wait for a third try. That is where customers quit.

How long will customers wait for a chatbot before they want a person?

Not long. Five9 surveyed 3,000 consumers in 2026. Of them, 92% give an AI agent under 10 minutes. Then they want a person. The top reason to leave the bot was repeated misunderstanding. That reason got 46%.

Do customers actually prefer humans over AI?

Most do, with one catch. Verint surveyed 5,000 U.S. consumers in 2026. Of them, 61% prefer a human agent. But 69% of that group would use self-service. The condition: it fully solves their problem. Customers want the problem solved.

What should happen during a chatbot handoff?

Three things go to the person. These are the full chat and the customer's name. The third is the trigger that fired. The bot tells the customer who is picking up. The customer never repeats the story.

Should my chatbot tell customers it's AI?

Yes. Five9 asked if people want to know when they talk to AI. Of them, 71% said it is very or extremely important. Customers could sometimes see a human option. Then trust in the bot more than doubled. Clear labels help the bot earn trust.

Get blog updates

Get an email when we publish a new post.

Ready to design AI experiences that actually work for your customers?

Book a Call