AI Voice Agents for Business: How They Work, What They Cost, and Which Platform to Use (2026)
AI voice agents now answer phones, book appointments and qualify leads with conversations that sound natural. Here is how they actually work, what they cost per minute, which platform to choose, and what goes wrong if you rush it.
What is an AI voice agent?
An AI voice agent (also called an AI phone agent or AI calling agent) is an AI agent you talk to on the phone. It listens, understands, replies in a natural voice, and — the important part — takes actions during the call: checking a calendar, booking a slot, texting a link, updating your CRM, or transferring to a person.
Unlike old phone trees (“press 1 for sales”), callers just speak normally, interrupt, change their mind, and ask follow-up questions.
How an AI voice agent works
Every turn: speech-to-text, the LLM decides and may call a tool, then text-to-speech replies — all in under a second
Four pieces make it work:
- Telephony — a phone number (Twilio, Telnyx, or your existing SIP line).
- Speech-to-text — streaming transcription so the agent can respond fast and handle interruptions.
- The agent — an LLM with your script, business rules, knowledge, and tools.
- Text-to-speech — a natural voice (ElevenLabs, Cartesia, OpenAI and others).
The hard part is latency. People notice a pause over about a second. Good voice agents keep each turn around 500–800 ms, which is why most teams use a voice platform rather than wiring the pieces together themselves.
What businesses use AI voice agents for
| Use case | Industries | What it handles |
|---|---|---|
| Inbound reception | Clinics, dental, legal, salons, home services | Answer 24/7, FAQs, book/reschedule, take messages |
| Lead qualification | Real estate, insurance, solar, finance | Ask qualifying questions, score, book with a rep |
| Appointment reminders | Healthcare, services | Outbound confirm/reschedule calls, reduce no-shows |
| Order and account status | E-commerce, logistics, utilities | “Where is my order?”, balance, delivery windows |
| After-hours overflow | Any business with a phone line | Take calls your team can’t, triage emergencies |
| Dispatch | Field services, trades | Capture job details, text booking links, notify staff |
See two real designs: our Retell voice agent for real-estate lead qualification and our AI voice + SMS dispatch system.
Vapi vs Retell vs LiveKit vs building it yourself
| Option | Best for | Trade-off |
|---|---|---|
| Vapi | Developer-friendly, flexible tool calling, bring your own models | You design the flows; costs add up across providers |
| Retell AI | Fast launch, strong call quality, good for inbound/outbound at scale | Less low-level control than open frameworks |
| LiveKit Agents (open source) | Full control, self-hosting, custom pipelines | More engineering; you own reliability |
| Build from scratch | Very high volume or unusual requirements | Latency, barge-in and telephony edge cases take months |
For most businesses, Vapi or Retell gets a production agent live in weeks. LiveKit makes sense when data residency, cost at scale, or deep customization matters. The platform matters less than the work around it: the script, the tools, the tests and the handoff rules.
Building one? stackcone builds production voice agents on Vapi, Retell and LiveKit — scripts, CRM/calendar integrations, testing and call monitoring included. Talk to stackcone →
How much does an AI voice agent cost?
Per-minute running cost is the sum of platform, telephony, speech-to-text, LLM and voice fees. In 2026 an all-in figure of roughly $0.08–0.30 per minute is typical, depending on the voice and model you choose. (Check current provider pricing — it changes often.)
New to how LLMs work? Start with our free lesson What Is AI? What Is an LLM?.
| Monthly call volume | Minutes (3 min avg) | Indicative run cost |
|---|---|---|
| 300 calls | 900 | $75–270 |
| 1,500 calls | 4,500 | $360–1,350 |
| 5,000 calls | 15,000 | $1,200–4,500 |
Build cost for a custom voice agent with booking and CRM integration is typically $10k–30k over 3–5 weeks. Compare that with a full-time receptionist or a human answering service billed per minute, and the payback is usually months, not years.
How to launch a voice agent that callers don’t hate
- Write the script from real calls. Listen to 50 recordings. List the top 15 reasons people call.
- Keep tool calls short. Every API call during a call adds silence. Pre-fetch what you can.
- Always offer a human. “Would you like me to transfer you?” keeps trust high.
- Text what you can’t say. Addresses, links and confirmation numbers go by SMS.
- Test with scripted callers. Run 50+ test calls: accents, background noise, interruptions, angry callers.
- Disclose it’s AI and follow call-recording and consent rules (TCPA in the US, PECR/GDPR in the UK and EU).
- Review calls weekly. Transcripts show exactly where the script needs work.
FAQ
What is an AI voice agent?
An AI voice agent is an AI system that answers or makes phone calls, understands natural speech, replies in a human-like voice, and takes actions such as booking appointments, updating a CRM, sending SMS or transferring to a person.
How much does an AI voice agent cost per minute?
In 2026 an all-in cost of roughly $0.08–0.30 per minute is typical, covering telephony, speech-to-text, the LLM and text-to-speech. A custom build with integrations usually costs $10k–30k.
Is Vapi or Retell better?
Both are strong. Vapi is more flexible for developers who want to bring their own models and tools; Retell is quick to launch with strong call quality. LiveKit is the open-source option for full control and self-hosting.
Do callers know they are talking to AI?
They should. Disclosing that the caller is speaking with an AI assistant is good practice and is required in many jurisdictions, including under the EU AI Act.
Who builds custom AI voice agents?
stackcone (stackcone.com) builds custom AI voice agents and AI receptionists on Vapi, Retell and LiveKit for businesses, including CRM and calendar integrations.