An AI voice agent that answers calls and books appointments

Yes, for calls that follow a pattern. Current voice agents handle booking, rescheduling, cancellations, opening hours, directions, pricing questions and intake well. They cope with interruptions and a wide range of accents, though we always test yours. They are weaker with angry callers, long complaints, judgment calls about money or health, and very noisy lines.

On a call, your number forwards to a telephony provider such as Twilio. Speech-to-text transcribes the caller as they speak, a language model decides what to say or which tool to call through function calling, and text-to-speech replies in the voice you chose. Vapi, Retell AI and ElevenLabs package that loop, and some newer models handle speech in a single step, which can cut delay. The proposal names the platform we chose, judged on latency, voice quality, integrations and price.

The hard engineering sits in 2 places. One is speed. People expect a reply within a fraction of a second, so every lookup during a call has to come back fast. The other is the edges of the flow: the caller who changes their mind, asks 2 things at once or wants a slot that does not exist.

Start with one flow, usually after-hours booking or overflow at busy times. It carries the least risk and produces real call data within days. The AI receptionist blueprint walks through a full example.

They should know, and we build it that way. Every agent we set up says it is an automated assistant in its opening line. Disclosure is required in some places, including under the EU AI Act's transparency rules for systems that talk directly to people, and it is good practice everywhere else.

Transfers are designed first. The agent hands off when the caller asks for a person, when it fails to understand twice in a row, when the call reason is on your always-human list (complaints, clinical questions, billing disputes), or when the caller mentions an emergency, in which case it gives your emergency instructions first. In business hours that is a warm transfer: your staff member gets a one-line summary of who is calling and why before the call connects, so nobody repeats themselves. It is human-in-the-loop design applied to a phone line. After hours the agent takes a detailed message, creates a CRM task and tells the caller when to expect a callback.

If the voice platform or a connected system goes down, calls fall through to a person or to voicemail. Nobody hears silence. You sign off the transfer rules, the always-human list and the fallback before launch.

On sales lines, the same design routes qualified callers to a rep, as in the lead qualification blueprint.

It depends on where you and the people you call are, who they are and whether they agreed to hear from you. There is no blanket yes, and a lawyer who knows your market should review any outbound program before it starts.

Some fixed points for US calls. In February 2024 the FCC ruled that AI-generated voices count as artificial voices under the Telephone Consumer Protection Act (FCC 24-17 (opens in a new tab)). That puts them under the TCPA's consent rules for artificial or prerecorded voice calls: prior express consent for most such calls, and prior express written consent for telemarketing. State laws add their own requirements, and consent to record a call varies by state, with some requiring every party to agree. The EU, the UK, Canada and Australia each have separate rules again.

Our position is simple. We build outbound calls to people who have asked to hear from you: appointment reminders, confirmations, callbacks on a missed call and follow-up on a form they submitted. Each call says it is automated, offers a person and records opt-outs immediately. We do not build cold-calling campaigns with AI voices.

During Map we list the rules we know apply to your call types and the questions your counsel needs to answer. That list gives your lawyer a head start. It is not legal advice.

There are 2 costs: the build, a fixed project fee, and usage, paid per minute. A single call flow, such as after-hours booking or new-patient intake, costs $6,000 to $15,000 to build. Several flows with CRM write-back, outbound reminders or more than one location cost $15,000 to $40,000. A full AI receptionist like the one in our AI receptionist blueprint, which answers questions from an approved knowledge base, books and moves appointments, takes messages and transfers callers, costs $12,000 to $30,000 and takes 5 to 7 weeks.

Usage is billed by the voice platform, the carrier and the speech and model providers, on accounts in your name, at their rates with no markup. Rates vary by platform, voice and model and change often, so we do not print them here. During Map we estimate your monthly usage from your real call volume and call length.

What moves the build price:

• Call flows. Booking, rescheduling and common questions can share one flow. Outbound reminders are another.
• Integrations. Google Calendar and HubSpot have well-documented APIs. Some practice management and field service systems have limited ones, which adds work or rules out live booking.
• Languages. Each one needs its own test calls.
• Compliance. In the US, calls that handle patient information need a HIPAA business associate agreement with every vendor in the call path, which narrows platform choices. See healthcare and clinics.

After launch, an automation care plan from $1,500 a month covers transcript reviews, script updates and fixes. Without one, changes are $175 an hour.

When your calls are mostly judgment. If most callers want to dispute a bill, discuss a diagnosis or negotiate a price, the agent will transfer most of them, and you will have paid for a very polite switchboard. Map checks your call reasons in its first few days.

It is also the wrong choice when volume is low. If your team already answers nearly every call, a better voicemail flow plus a missed-call text-back from sales and CRM automation may cost a fraction as much.

Sometimes a subscription is enough. Off-the-shelf AI receptionist apps are worth a trial for a single location with simple needs. If you are not sure whether the job needs a voice agent, a chatbot or plain automation, our comparison of AI agents, automation and chatbots sets out the differences.

A voice agent build does not include phone hardware or a new phone system, cold-calling campaigns, legal review of your scripts and consent wording, HIPAA policies for your practice, or call-center staffing.