AI to human handoff: when your AI caller should pass the call to a person

An AI caller should hand over to a human when the caller asks for a person, when they raise a complaint, when the question goes beyond what the agent can answer accurately, and when the deal is valuable or sensitive enough that a mistake would be expensive. A good AI to human handoff also passes on what has already been said, so the caller never has to start again. And if someone asks who is calling and how to complain, they need a clear answer straight away. That last part is a legal requirement, not just good service.

This guide is for Australian teams already running AI on their phones, or about to. It covers the triggers worth building in, how to hand over smoothly, and the questions to ask before you go live.

Why the hand-off matters as much as the script

Most of the effort in setting up an AI voice agent goes into the happy path: the greeting, the qualifying questions, the booking. That's reasonable, because most calls follow it. But callers judge your business by the calls that go off-script. A prospect who asks a tricky question and gets a vague loop of answers, or who asks for a person and gets ignored, remembers it.

A well-designed hand-off does two things. It protects the relationship when the AI is out of its depth. It also lets the AI handle routine work with confidence, because you know there's a safety net for everything else.

The triggers that should send a call to a person

Write your hand-off triggers down before launch. Vague instructions like "escalate when needed" leave too much to interpretation. These are the categories most teams need.

The caller asks for a human

This is the simplest trigger and the one that most often goes wrong. If someone says "can I talk to a real person?", the agent shouldn't try to talk them out of it or ask them to rephrase. It should acknowledge the request and move to a transfer or a booked callback.

Include the many ways people phrase this request: "is this a robot?", "put me through to someone", "I'd rather speak to your manager". Test them all.

A complaint or clear frustration

When a caller is unhappy, about the call itself, a past experience or your business generally, a person should take over. AI can acknowledge the issue and gather the basics, but a complaint usually needs judgement, empathy and the authority to fix something. Signs to watch for include raised language, repeated objections, mentions of a previous bad experience and phrases like "this is ridiculous" or "I've already told you".

Questions beyond what the agent knows

Your agent should only answer from approved information: pricing ranges you've signed off, service areas, availability, standard terms. When a question falls outside that, the right move is to say so and hand over, not improvise. Common examples:

• Custom quotes or exceptions to standard pricing
• Technical or clinical questions that need a qualified person
• Contract terms, refunds or warranty disputes
• Anything touching legal, financial or health advice
• Questions about an existing account the agent can't see

A confident wrong answer does more damage than "that's a great question for our specialist, let me get them to you."

High-value or strategically important deals

Some calls are worth a person's time even when the AI could technically handle them. Set rules based on what matters to your business: a large project size, a commercial client, a referral from a key partner, or a prospect who mentions they're comparing options and ready to decide this week. For these, the AI's job is to qualify, capture the details and get a senior salesperson involved quickly.

Signs of vulnerability or confusion

If a caller seems confused about who is calling or why, mentions difficult circumstances, or appears unable to make an informed decision, stop selling and hand over or end the call politely. This protects the caller and your business.

Contact and complaint details: what you must give when asked

Australian telemarketing rules speak directly to this. Under the Telecommunications (Telemarketing and Research Calls) Industry Standard 2017, if asked, a telemarketing caller must immediately give the contact details of their employer and of the person who caused the call, plus the name and contact details of anyone responsible for handling inquiries or complaints about the caller, the employer or that person.

For AI calls there's a specific point to note. Where a call involves recorded or synthetic voices, the Standard says the caller must provide a way during the call for the recipient to request the caller's contact and complaint-handling details, for example by pressing a button to talk to an operator.

In practice, that means:

• Your agent should be able to give your business contact details and complaints contact on request, without hesitation
• The complaints contact should be a real, monitored channel, not a generic inbox nobody reads
• There should be a working way for the caller to request those details during the call, whether that's a transfer, a keypress or a spoken request the agent recognises
• Test this path as carefully as your booking flow

The opening of the call matters too. The same Standard requires a telemarketing caller, as soon as the call starts, to give their given name (unless the call is made solely using a recorded or synthetic voice), the business name of their employer, the name of the person causing the call to be made, and the purpose of the call. The ACCC also notes that the consumer cooling-off period extends to 3 months if the salesperson calls outside the allowed hours, doesn't give their name or the name and address of the business they represent, or doesn't say why they're calling. Getting the identification right protects the sale as well as the caller.

How these rules apply to your particular calls depends on your circumstances, so check with the ACMA or a lawyer if you're unsure.

How to hand over without making the caller repeat themselves

Nothing undoes the goodwill of a hand-off faster than a person picking up and asking "so, how can I help?" The caller has just explained everything. Here's how to avoid it.

Pass a written summary, not just a recording

Before the transfer connects, your team member should see a short summary on screen or in their notification: who the caller is, what they want, what's been covered, what triggered the hand-off and anything they've already been told. A full transcript is useful for reference, but nobody can read it in the seconds before picking up. The summary is what gets used.

A good summary covers:

• Caller name and the enquiry they made
• What they're looking for, in their words where possible
• Answers to qualifying questions already asked
• The reason for the hand-off, such as "asked about a custom quote" or "unhappy about a previous delay"
• Any promises the AI made, such as "told them someone would call back today"

Choose between warm transfer and booked callback

A warm transfer, where the person joins the live call, is best for complaints, people asking for a human and hot opportunities. It only works if someone is actually available. A transfer that rings out is worse than none.

A booked callback suits complex questions that need preparation, or times when your team is busy. The AI should offer a specific time, confirm the best number and explain who will call. Then the callback has to happen when promised.

Many teams use both: try a warm transfer during staffed hours and fall back to a booked callback otherwise.

Open the human side of the call properly

Train your team to open with what they already know: "Hi Sam, I understand you're after a quote for a commercial fit-out and had a question about timing." This tells the caller they've been heard and saves several minutes of repetition.

Tell the caller what happens next

Before the hand-off, the AI should say plainly what's about to happen: "I'm going to put you through to Priya from our sales team. I've passed on what you've told me so you won't need to repeat it." Setting that expectation reduces anxiety and drop-offs.

When no one is available

Plan for the times nobody can take the call. The AI should never leave a caller in silence or on endless hold. Instead it should:

• Explain that the right person isn't available right now
• Offer a specific callback time or let the caller choose one
• Still provide contact and complaint details if asked
• Log the request so it lands in front of a person with the summary attached

Review these unanswered hand-offs regularly. If they pile up at certain times, adjust your rostering or the hours your agent calls.

Questions to ask your AI voice provider

Before you rely on any platform, ask:

• Can I define my own hand-off triggers, including specific phrases and deal criteria?
• Does the agent always recognise a request for a human, however it's phrased?
• What does my team see at the moment of transfer, and how quickly?
• How does the agent give contact and complaint details on request?
• What happens if a transfer isn't answered?
• Can I review transcripts of every hand-off to improve the rules?

Putting it into practice

Start with a short list of triggers, test each one with real people role-playing difficult callers, and review hand-off transcripts every week for the first stretch after launch. Tighten the rules where the AI held on too long, and loosen them where it handed over calls it could have finished.

At Dialessa, this is how we think about the job: our AI voice agent calls new enquiries within a minute, books the appointment and follows up past leads whose consent is still current. When a conversation needs a person, the aim is to get them there with the full context already in hand.

The best AI calling setups aren't the ones that never hand over. They're the ones that hand over at the right moment, so the caller feels looked after the whole way through.