An AI voice agent pays off for a service business when inbound calls are frequent, repetitive and follow a predictable script, such as booking, rescheduling, opening hours or order status, and when the agent can finish the task on the call or hand it to a person with the context. If most calls need judgement, start elsewhere.

Picture a plumbing company with two office staff. In the morning rush the phones ring while both are already on calls. Some callers leave a voicemail; others ring the next company on the list. Most of those calls are the same four questions: can someone come today, what does a call-out cost, can I move my appointment, is the engineer on the way. The question is not whether a voice bot sounds human. It is whether it can answer those four questions correctly and book the job.

What is an AI voice agent?

An AI voice agent is software that answers or places phone calls, understands what the caller says, and replies in a synthetic voice while taking actions in your systems. Under the hood it chains speech-to-text, a large language model and text-to-speech, connected to a phone line.

It differs from an old interactive voice response (IVR) menu, the "press 1 for bookings" kind. An IVR routes callers through fixed options. A voice agent holds a conversation, asks follow-up questions and can call tools: check the calendar, create a booking in the field-service system, look up an invoice, or transfer the call. Platforms such as Retell AI and Vapi provide the voice and telephony layer; you supply the script, the knowledge and the integrations.

Which calls should a voice agent take?

A voice agent should take calls that are high in volume, low in judgement and end in a clear action. Calls that need negotiation, empathy in a difficult moment or a decision outside the rules should go to a person.

Good first candidates

• After-hours and overflow answering: take the details, book a slot or promise a callback.

• Booking and rescheduling: where the rules (service area, slot length, lead time) can be written down.

• Status questions: "is my technician on the way", "has my order shipped", read from a system of record.

• Simple qualification: collect the job type, address and urgency before a person calls back.

Keep with people

• Complaints, refunds and disputes.

• Emergencies where safety advice matters.

• Quotes that depend on an on-site judgement.

• Anyone who asks for a person.

Before you build anything, listen to a sample of real calls and sort them into these buckets. If the first bucket is small, a voice agent will mostly transfer calls, and you pay for both.

What does an AI voice agent cost to run?

AI voice platforms usually charge per minute of call time, and the per-minute price is the sum of several parts: the platform fee, speech-to-text, the language model, the voice, and telephony. The model and voice you choose move the total most.

As of 11 October 2026, the vendors' own pricing pages show:

• Retell AI: charges per minute, built from components. The page lists "$0.07-$0.31 / min" for voice agents depending on configuration, with voice infrastructure at $0.055 per minute plus voice, model and telephony.

• Vapi: charges a platform fee per minute plus providers. The page lists a $0.05 per minute hosting fee, with speech-to-text, model, voice and telephony providers passed through "at cost".

Prices change, so check both pages on the day you budget. Also count the costs that are not per minute: building and testing the script, integrating your calendar or CRM, and someone reviewing calls every week.

How do you know if a voice bot pays off?

A voice bot pays off when the cost of the minutes it handles, plus the work to build and maintain it, is lower than what those calls cost you today, including the calls you currently miss. Work it out per call type, not for the whole phone line.

A simple way to check

1. Count the calls in each candidate bucket for a normal month, including missed and after-hours calls.

2. Measure the current cost of each: staff time per call, and the value of a booking you lose when nobody answers.

3. Estimate the agent's cost per call: average call length times the per-minute price from the pricing pages above.

4. Estimate the share it can finish without a transfer. Test this on real call recordings, not on a guess.

5. Add the fixed costs of setup, integrations and weekly review.

6. Compare, and start with the one bucket where the gap is clearest.

If you can't fill in steps 1 and 2 from your own phone system, fix the call tracking first. Any business case without those numbers is a guess.

What rules apply to AI voice calls?

The rules depend on where your callers are, but two apply widely: callers must know they are talking to an AI, and automated outbound calls need consent. This is a summary, not legal advice; check the rules for your market with a lawyer.

• United States: On 8 February 2024 the Federal Communications Commission ruled that "calls made with AI-generated voices are 'artificial'" under the Telephone Consumer Protection Act (TCPA). Outbound calls using an AI voice therefore fall under the TCPA's consent rules for artificial or prerecorded voices.

• European Union: EU AI Act Article 50 requires that AI systems meant to interact directly with people are built so that people "are informed that they are interacting with an AI system", unless that is obvious. Our post on Article 50 for chatbots and AI content covers what it means in practice.

• Call recording: most voice platforms record and transcribe calls by default. Consent rules for recording differ by country and, in the US, by state, so say at the start of each call that it is recorded and why.

Inbound agents are the easier start: the caller chose to call you, and a one-line disclosure ("You're speaking with our AI assistant; I can book, reschedule or connect you to the team") covers the transparency part.

What happens when the voice agent can't help?

When the voice agent can't help, it should transfer the call to a person, with a summary, and never leave the caller stuck in a loop. The handoff decides how the whole experience feels.

Both platforms support this. Retell's transfer call tool offers cold, warm and "agentic warm" transfers, where the agent can brief the person before connecting the caller. Vapi's transfer call tool supports blind transfers and several warm-transfer modes, including passing a summary. Design for these cases:

• The caller asks for a person: transfer at once, without arguing.

• The agent fails to understand twice: transfer, don't ask a third time.

• Nobody is available: take a callback request with a time window and create a task in the CRM.

• After every call: write the outcome and transcript to the CRM so the team sees what the agent did.

How we approach it at Vizio AI

We start from the calls, not the bot. With the client we pull a sample of recent recordings, sort them into the buckets above, and pick one call type with a clear action and enough volume. We write the script and the transfer rules together with the people who answer the phone today, because they know the edge cases.

Then we connect the agent to the systems it needs (calendar, field-service tool or CRM) through an automation layer the team can maintain; our comparison of n8n, Zapier and Make explains how we choose one. Before go-live we replay real calls against the agent, check every write it makes, and agree a weekly review of transcripts. Only when one call type works do we add the next. This sits in our DX Studio work on operations, alongside the scheduling side covered in How to manage automated operations calendars.

Sources

• Federal Communications Commission, FCC Makes AI-Generated Voices in Robocalls Illegal (8 Feb 2024): https://www.fcc.gov/document/fcc-makes-ai-generated-voices-robocalls-illegal

• Regulation (EU) 2024/1689 (Artificial Intelligence Act), Article 50: https://eur-lex.europa.eu/eli/reg/2024/1689/oj

• Retell AI, Pricing (retrieved 11 Oct 2026): https://www.retellai.com/pricing

• Vapi, Pricing (retrieved 11 Oct 2026): https://vapi.ai/pricing

• Retell AI Docs, Transfer call tool: https://docs.retellai.com/build/single-multi-prompt/transfer-call

• Vapi Docs, Transfer Call tool: https://docs.vapi.ai/call-forwarding