When Should an AI Hand Off to a Person? A Practical Guide

An AI should hand off when the caller asks for a person, when the stakes are high, or when it cannot finish the job. No single law or vendor sets a universal list, so you need to write your own rules before going live.

The triggers below are our recommended starting point, not a legal standard. Vendor features were checked against each vendor's documentation on 9 November 2026. This article is part of our AI Front Desk guide.

Seven moments to hand off

1. The caller asks for a person. Make this the simplest rule you have. Don't make callers repeat themselves or try a few more questions first.

2. There is an emergency or crisis.

Don't let an AI assess it. If someone is in danger, tell the caller to hang up and dial 911. The 988 Lifeline says: "The 988 Lifeline is available 24/7/365."

3. The caller needs professional judgment. Medical, legal and financial advice should come from qualified people.

Florida's Ethics Opinion 24-1 is one advisory example for lawyers. It says nonlawyer intake staff should refer legal questions back to the lawyer, and it warns about chatbots that may give legal advice.

4. The caller is upset. Repeated complaints, raised voices or threats to leave are good reasons to move the call.

5. The AI can't finish the task. Examples include a failed lookup, a calendar error or a request outside its instructions. Set a limit, such as two failed attempts, then transfer.

6. Money or account changes are involved. Payment disputes, refunds and cancellations carry more risk when something goes wrong. Have people handle these unless you have tested a narrow, approved flow.

7. The caller asks whether it's AI. Answer truthfully, then offer a person.

Utah Code 13-77-103 requires disclosure in consumer transactions if the person clearly and unambiguously asks whether they are dealing with AI. It adds stricter rules for regulated occupations in high-risk interactions.

Be honest before the handoff too

Disclosure rules vary by state, and they don't wait for a transfer. Maine's law bars using a chatbot in trade in a way that may mislead a consumer about talking to a human, unless clearly notified. Its definition covers "textual or aural communications."

Florida's opinion says lawyers must tell prospective clients they are communicating with an AI program. It also says the lawyer remains ultimately responsible if the chatbot gives misleading information. The opinion is advisory and applies to Florida lawyers.

This is general information, not legal advice. Ask counsel which rules apply to your business and your states.

Cold, warm and briefed transfers

The transfer type decides how much your staff knows when they pick up.

Here is how three developer platforms describe this:

• Retell's transfer docs explain cold transfer this way: "The call goes straight to the destination and your agent drops off right away." For warm transfer, they say: "Your agent stays on the line after dialing the destination."
• Vapi's assistant-based warm transfer puts the customer on hold while a transfer assistant calls the destination. Vapi says: "The caller stays on hold and doesn't hear the operator conversation."
• Bland's warm transfer has the AI brief the human agent on a separate call, then merge the calls. It is off by default.

Packaged products can say less about how it works. Weave's AI Receptionist page says: "When calls need a human touch, AI Receptionist hands them off for a smooth experience."

Plan for the transfer that fails

A handoff only helps if somebody picks up. Decide in advance what happens when nobody does.

Each platform handles failure in its own way:

• Retell. With an agentic warm transfer, you choose what happens at timeout. If you pick cancel, the caller goes back to the main agent.
• Vapi. If the transfer fails and you set no message, the original assistant says "Could not transfer the call." Vapi also says voicemail isn't detected automatically with a transfer assistant, so you must prompt for it.
• Bland. The timeout runs from 30 to 3,600 seconds. Bland says: "After the timeout expires, the system gives up on the transfer and moves to your configured voicemail handling."

A good failure plan has four parts:

1. Tell the caller what is happening, and never leave them in silence.
2. Offer a callback at a specific time.
3. Capture the caller's name, number and reason for calling.
4. Send that to a person who owns the follow-up.

If that follow-up is a call placed by AI, remember it is an outbound call with different consent rules.

Give your staff the context

A transfer fails in a quieter way when the caller has to start over. Pass along a few facts: the caller's name, the reason for the call and what the AI already tried.

Check that your setup really does this. Vapi's page on its Transfer Call tool says the destination "doesn't receive the transcript, variables, or assistant context." Add a briefing so your staff don't start from nothing.

What transfers cost

Transfer time can be billed differently from AI time, so include it in your estimate. These were the published terms on 9 November 2026.

• Retell. After the caller connects and the agent drops off, its docs say the AI agent fee stops. The telephony fee continues for the rest of the call.
• Bland. Its pricing page lists transfer minutes at $0.05 on the Start plan and $0.04 on Build.

Our AI receptionist pricing guide covers the wider picture.

Test before callers do

Before going live, run each transfer path yourself. Try a transfer during business hours, after hours, with a busy line and with no answer.

Then review real transcripts in the first weeks. Look for calls where the AI should have transferred and didn't, and for transfers that didn't need to happen. Adjust your rules based on what you find.

If you need these transfer rules connected to your phone system or CRM, our AI agent development team can scope that work.

Frequently asked questions