Voice agents · September 14, 2026

What should an AI voice agent do when a caller needs a human?

Cornerstone can route a caller to your owner or office. A useful handoff also needs a plan for what happens next.

By

Picture a customer calling your business about a charge they don't understand. Your voice agent answers, gets their name, and starts asking about appointments. The customer says, "Can I talk to somebody?"

That should be enough. They shouldn't have to argue with your phone system to reach a person.

Cornerstone, the platform I use, can route the call to your business owner or office when someone asks for a human. The part we need to work out together is where that call goes, when somebody will be there to answer, and what happens if they can't pick up.

I want that worked out before the agent answers your first customer.

Human handoff flow: a caller asks for a person or needs human judgment; route to the owner or office; if someone answers, pass along the reason; if nobody answers, explain the next step and assign a callback.
A practical handoff plan. Cornerstone supports routing to the owner or office; the callback process and information passed along need to be configured for your business.

1. When should the agent get a person?

An explicit request is the easy one. "Let me talk to the owner," "I need someone in the office," and "Can I speak to a real person?" should lead to the human handoff you've agreed on. Asking the same question three different ways ought to be a test you run, not something your customer has to do.

Some calls need a person even when the caller doesn't ask. I'd set that boundary around decisions the agent isn't authorized to make: a disputed bill, a refund exception, a complaint about the work, or a promise about when a technician can arrive that the schedule doesn't support.

The agent can collect the reason for the call. The person responsible for the decision should make it. A friendly voice doesn't give the software authority to approve a refund.

I'd also give it a way out when the conversation gets stuck. If it keeps misunderstanding the customer or can't find an answer in your approved information, it should offer a person instead of repeating the same response.

2. What should the caller hear?

Keep it short and tell them what's about to happen. Here's sample wording for a call during staffed hours:

"I can try the office for you. I'll put you on hold while I connect the call."

That sets an honest expectation. Saying "The owner is available" would require knowing that the owner is available. A phone number in a routing rule doesn't tell you whether somebody is standing next to it.

Ask for the minimum information needed to help. A brief reason for the call and a confirmed callback number can be useful if the connection fails. I wouldn't make someone complete your entire new-customer intake before letting them speak to the office.

3. Who should receive the call?

If you're the owner, sending everything to your cell phone may sound simplest. Then you're under a sink or talking with another customer, and the call rings through to voicemail anyway.

Choose the destination around how your business works. During office hours, that might be whoever handles scheduling and customer questions. A complaint that needs your decision might go to you. Outside staffed hours, the agent should follow the after-hours plan you approved.

Write down the destination, the hours it should be used, and the fallback. If your office number sends unanswered calls back to the agent, test that too. You don't want the customer riding in a circle.

For more on that part of the setup, I wrote about what should happen to a call after your business closes.

4. Will the customer have to start over?

Try to pass along what the customer already explained. At a minimum, the person taking over needs the reason for the call and any next step the agent has already discussed.

Some systems offer a warm transfer, which can brief the person receiving the call before connecting the customer. Retell's documentation, for example, describes private messages to the recipient and introductions after connection. Those are options to check in your specific setup; routing a call by itself doesn't establish that those features are enabled.

If a spoken introduction isn't available, a short note in the place your team already checks can help. I'd rather give the office five useful lines than expect somebody to read a full transcript while the phone rings.

Example callback noteIllustrative details

Give the next person enough to act.

Reason
Customer wants an explanation of a charge on the last invoice.
Contact
Name and callback number confirmed with the caller.
What happened
Office transfer attempted. Nobody answered.
Next step
Office manager to review the invoice and return the call.
Expectation
Caller was told the office will follow up during its next staffed period.
This is an example of the information to pass along, not a screenshot of a built-in Cornerstone feature.

5. What if nobody answers?

"But Jason, the whole reason I need an agent is that we're busy."

Exactly. That's why an unanswered transfer needs its own plan. The caller should hear that nobody picked up and get a useful next step. Depending on your setup, that might mean taking a callback request, trying an approved backup number, or explaining when the office opens.

Use a callback window your team can meet. If you haven't committed to returning calls within ten minutes, the agent shouldn't invent that promise. It can explain your actual office hours and how the request will be handled.

Then assign the request to somebody. A notification sent to an inbox that everyone assumes someone else is watching is still an unanswered customer.

I also want the agent to distinguish between "I'll send the office a note" and "I've sent the office a note." It should only claim the action is complete after the system confirms it. If that step fails, your team needs a way to find out.

6. How do you know the handoff works?

Call the actual business number and run through the situations your customers will encounter. Retell's transfer troubleshooting guide points out that its call transfers require a phone call, rather than a web call. Whatever platform you use, test the complete phone route before relying on a browser demo.

  • Ask for the owner, then ask for the office using different wording.
  • Raise a billing question the agent isn't authorized to resolve.
  • Let the receiving phone ring without answering, then test what happens when voicemail picks up.
  • Call outside the hours you've set for live transfers.
  • Check whether the right person receives the callback request with the correct number and reason.

Listen to what the caller hears throughout the attempt. Check for a long unexplained hold, a disconnected call, or a promise your staff never receives. Fix the failed path and run it again.

Once it's live, review missed transfers and unfinished callbacks. An agent can answer the phone successfully while the customer is still waiting for help. Your follow-up record should make that visible.

A person should be easy to reach

My standard is pretty simple: when a caller needs a person, your system should help them reach the right one. If that person isn't available, the customer should understand what happens next and your team should know who owes the callback.

Cornerstone can handle the routing. We still need to agree on the destination and build a follow-up process your team can keep up with. That's the work that makes the handoff useful.

Where should your calls go?

Bring one call situation that's difficult for your business to a 20-minute video call. I'll help you map when the agent should handle it, when a person should step in, and what should happen if nobody answers.

Book a 20-minute video call See the AI voice agent service