AI voice agent pricing in India: what to include in the cost per useful call
Per-minute prices tell only part of the story. Compare telephony, speech, model usage, failed calls, recordings, and the work your team still needs to do.
Vistrow Voice team
Product, speech, and customer experience
Ask what a quoted minute actually includes
Two per-minute quotes are not comparable until you know the billing boundaries. Does the price include the phone connection, speech recognition, speech synthesis, language-model usage, storage, and platform fees? Are unanswered calls or setup time billed? Ask for a worked invoice example based on your expected traffic.
Check minimum commitments, concurrency limits, rounding, currency, taxes, and support separately. A low headline rate may still be a good deal, but it should not be compared with an all-in rate as if both cover the same service.
Once you have a cost per useful call, compare it with the current Vistrow Voice pricing page using the same traffic assumptions. The latency measurement guide helps you check whether a lower-cost configuration still gives callers a responsive experience.
Use a simple cost worksheet
Estimate attempted calls, connected calls, average connected duration, and monthly fixed fees. Add the usage charges included in the contract and any provider charges billed separately. Keep retry traffic and testing traffic visible rather than hiding them inside an optimistic average.
For illustration only: if 1,000 connected calls average three minutes, that is 3,000 connected minutes before retries or other charges. At a hypothetical all-in rate of ₹4 per minute, the usage component is ₹12,000. This is arithmetic, not a Vistrow quote or a market price. Replace the assumptions with an actual proposal.
- Connected minutes × applicable usage rate.
- Phone numbers, platform minimums, and other fixed fees.
- Additional storage, model, or integration charges if excluded.
- Retries, testing, and unsuccessful attempts where billable.
- Staff review and follow-up time.
Cheap turns can produce expensive conversations
A cheaper model that needs repeated clarification may lengthen calls or collect the wrong details. A more expensive voice can still be worthwhile if callers understand it better, but that should be tested rather than assumed. Compare task completion and field accuracy alongside minutes and cost.
Define “useful” before calculating cost per outcome. For a callback flow, it might mean a valid contact, a clear reason, and an agreed next action. Do not label every connected call a success. Divide total operating cost by verified useful outcomes, and report the underlying sample size.
Run a limited pilot before committing volume
Use one campaign or inbound call reason. Track actual duration, abandoned calls, human escalations, and correction work. Ask the vendor to explain discrepancies between your records and the invoice.
Provider rates change, so use current written pricing and keep its date with the worksheet. Once the pilot shows reliable outcomes, project a conservative monthly range rather than a single best-case saving.