What an AI voice agent actually costs to run in production
Vendor pricing pages quote per-minute rates that exclude telephony, orchestration and the human review you will need. Here is the full number.
The quoted rate is not the cost
Model inference is one line item. A production deployment also pays for telephony minutes, speech-to-text, text-to-speech, orchestration hosting and CRM integration calls. Realistically, all-in cost lands between ₹6 and ₹12 per connected minute for Indian-language deployments.
Where the money actually goes
Text-to-speech is usually the largest component for natural-sounding Indian voices, followed by telephony. Model inference is frequently the smallest line, which surprises most buyers.
- Telephony: ₹0.60–₹1.20 per minute
- Speech-to-text: ₹0.80–₹1.50 per minute
- Text-to-speech: ₹2–₹4 per minute for premium voices
- Model inference: ₹1–₹3 per minute depending on turn count
- Orchestration and hosting: ₹15,000–₹40,000 per month fixed
Compare against the real alternative
A staffed calling desk in Gujarat costs roughly ₹25,000–₹35,000 per agent per month for around 120 productive call-minutes a day. That works out near ₹11–₹14 per minute before supervision, attrition and training.
The agent is therefore roughly cost-neutral to modestly cheaper per minute — and dramatically cheaper at peak, because it scales to unlimited concurrent calls with no hiring lead time.
Budget for review
Plan for a human to review a sample of calls weekly for the first three months. Skipping this is how deployments quietly degrade. Budget four to six hours a week initially, tapering after the first quarter.
Key takeaway
Expect ₹6–₹12 per connected minute all-in. The economics work on availability and peak scaling, not on undercutting a human on unit cost.