What an AI voice agent actually costs to run
Vendors quote a per-minute rate. The real number includes telephony, transcription, the model, and the calls it hands to a human. Here is the full arithmetic.
The question we get asked most often about voice agents is what they cost. The honest answer is that the headline per-minute rate is roughly half the real number, and the half nobody quotes is the half that varies most between businesses.
The four things you pay for
Every voice agent call bills across four separate layers, and they stack:
- Telephony — carrying the call itself, billed per minute, plus the number rental
- Speech to text — turning the caller's audio into text the model can read
- The language model — reasoning over the conversation and deciding what to do
- Text to speech — the voice your caller hears, priced per character or per second
As a rule of thumb in mid-2026, a straightforward reception-style call lands somewhere between eight and twenty cents a minute all in, depending mostly on which voice and which model you choose. A premium voice can cost more than the reasoning does.
The costs that are not per-minute
This is where budgets go wrong. Three line items sit outside the per-minute rate and usually dominate the first year.
Escalations
Some percentage of calls will be handed to a person. That is a feature, not a failure — but it means you are paying for both the agent's attempt and the human's time. Model your cost per resolved call, not your cost per minute, or you will flatter the numbers.
Integration work
An agent that cannot see your calendar or write to your CRM is an expensive answering machine. The build cost is dominated by wiring it into the systems where the work actually lands, and that is a one-off engineering project, not a subscription.
Evaluation and tuning
You need to listen to real calls, find the ones that went wrong, and fix them. Budget ongoing time for this. Teams that skip it end up with an agent that quietly degrades as the business changes around it.
When the arithmetic works
Voice agents pay for themselves fastest where call volume is high, calls are short and repetitive, and the cost of a missed call is concrete. Appointment booking, opening hours, order status, reservation changes, out-of-hours triage.
They pay off slowest where every call is different, emotionally charged, or where the caller has already tried the website and failed. Those callers want a person, and giving them a machine costs you more than the call.
The strongest case for a voice agent is rarely cost per call. It is the calls you currently never answer.
For most of the businesses we work with, the compelling number was not the saving on staffed hours. It was the third of inbound calls that previously rang out after six and were never returned.