What an AI voice agent actually costs to run
Telephony, speech, the language model and support. A plain breakdown of the monthly costs behind a voice agent, and the choices that move them most.
When companies ask what a voice agent costs, they usually mean the build. But the number that decides whether it pays off is the monthly running cost per call. That cost has four parts.
1. Telephony
Every call runs over a phone line, through a provider such as Exotel, Knowlarity, Plivo or Twilio. You pay per minute, plus a rental for numbers. This part is predictable and is often similar to what you already pay for your human team’s calls.
2. Speech in and speech out
The agent has to turn the caller’s voice into text, and its reply back into voice. Both are charged by the minute or by the character. Indian languages and mixed Hindi-English speech need models that handle them well, and the better ones cost more. This is usually the biggest single line.
3. The language model
The model that decides what to say is charged by the amount of text it reads and writes. A short confirmation call costs very little. A long, open conversation that looks up records several times costs more. Keeping prompts tight and using smaller models for simple steps makes a large difference here.
4. Monitoring and improvement
Someone has to listen to a sample of calls, fix answers that went wrong and update the script when your offers change. This is a fixed monthly cost rather than a per-call one.
What moves the cost most
- Call length. A focused two-minute call costs a fraction of an open-ended eight-minute one. Design the conversation around one goal.
- Language mix. Supporting more languages is worth it, but test whether each one earns its cost.
- Hand-off timing. Handing a difficult caller to a person early is cheaper, and better for the caller, than letting the agent struggle.
- Volume. Most providers get cheaper per minute as volume grows. Negotiate once you have a few months of real numbers.
How to compare it fairly
Compare the cost per completed outcome, such as a booked visit or a confirmed payment, not the cost per minute. A voice agent that calls every lead within five minutes, at any hour, often wins on outcomes even where it doesn’t win on minutes.
We share a full cost estimate, line by line, after the discovery call, based on your expected call volume and languages.