Voice AI Agent Cost Per Minute: The Real Math
Vendors love quoting a single "per minute" price for voice agents. That number hides five separate meters, and which one dominates depends on your call pattern. Here is how we break it down for clients before they commit.
- Published
The five meters running on every call
A voice agent call is a pipeline, and every stage bills separately. Knowing the stages lets you see which lever actually moves your bill.
- Telephony — the phone number and the per-minute carrier charge for inbound or outbound calls.
- Speech-to-text — streaming transcription of the caller, billed per audio minute.
- The language model — billed per token, so it scales with how much the agent says and how much context it re-reads each turn.
- Text-to-speech — the synthetic voice, billed per character or per minute of generated audio.
- Orchestration and hosting — the server that glues it together, plus logging and recordings storage.
On a typical booking call, speech-to-text and text-to-speech together are usually the largest share. The LLM becomes the largest share only when prompts are long and the conversation runs many turns — which is exactly what prompt caching and tighter prompts fix.
Typical all-in running cost
| Setup | What it uses | Relative cost | When it fits |
|---|---|---|---|
| Budget stack | Standard STT, compact LLM, standard voice | Lowest | High-volume reminders, simple FAQs |
| Balanced stack | Low-latency STT, mid-tier LLM, natural voice | Medium | Bookings, qualification, most clinics |
| Premium stack | Best STT, frontier LLM, cloned brand voice | Highest | Sales calls where tone converts |
Provider prices change several times a year, so we price every project against the current rate cards at kickoff and give you a per-minute figure in the proposal. What does not change is the shape: the balanced stack is right for most businesses, and the premium stack only pays back on calls where the voice itself sells.
The build cost sits on top
Running cost is only half of it. Building the agent — call flows, calendar or CRM integration, the eval suite, human hand-off and consent handling — is a one-time project. Our fixed-price ranges are $8k–$18k for a booking line, $12k–$25k for a sales line that qualifies and routes hot leads, and $20k–$40k for combined inbound and outbound with reminders.
Integrations are the swing factor. A clean calendar API is a day of work; a legacy practice-management system with no API can be two weeks.
When it beats a human receptionist
- After-hours and overflow calls: the agent answers calls that would otherwise go to voicemail — which is pure recovered revenue.
- Repetitive, structured calls: bookings, reschedules, order status, reminders.
- Volume spikes: a campaign or a festival weekend does not need extra hiring.
It does not beat a skilled human on complaints, negotiations or emotionally loaded calls. The right design routes those to a person within seconds — the agent handles the routine so your people handle the rest.
Frequently asked questions
Is a cheaper per-minute rate always better?
No. A cheaper speech model that mishears names or dates causes failed bookings, and one failed booking costs more than thousands of minutes of the better model.
Can we cap the monthly bill?
Yes. We set per-call time limits, daily spend alerts and a hard monthly ceiling at the orchestration layer, so a runaway loop can never surprise you.
Do recordings add cost?
Storage is cheap, but retention has compliance implications. We store what your consent notice covers and delete on schedule.
