📞 AI voice calling

What an AI voice call really costs per minute in India

Updated 2026-09-19 · 4 min read · By the Convo360 team

In short

An AI voice call's cost is not one number — it blends four separate components: telephony minutes, speech-to-text, the LLM generating replies, and text-to-speech. On a BYOK platform like Convo360 you pay each provider directly at their own rates with no platform markup, so your actual per-minute cost depends entirely on which providers and models you choose, not on Convo360's plan price.

Key takeaways

A voice call's cost is a stack of four components, not one line item: telephony, STT, LLM tokens, and TTS.
On BYOK, Convo360 adds zero markup on minutes or tokens — you pay each provider at their own rate.
Incoming and outbound call charges depend on your telephony provider and plan; check the provider's current price list rather than assuming incoming is free.
Shorter, well-designed scripts reduce LLM token usage and TTS duration, which lowers cost per call directly.
Comparing platforms by 'per-minute markup' vs BYOK is more useful than comparing headline plan prices.

Why 'cost per minute' is really four costs

When people ask what an AI voice call costs, they usually picture one number, like a mobile recharge plan. In practice, an AI-handled call blends four separate cost components, each billed by a different provider:

Telephony: the actual phone minute, billed by your telephony provider (Exotel, Plivo, Twilio or SIP).
Speech-to-text (STT): converting the caller's voice into text the AI can process.
LLM: the language model generating the agent's replies, billed by tokens processed.
Text-to-speech (TTS): converting the AI's reply back into spoken audio in the chosen voice.

Each of these is a separate line item from a separate provider. There is no single 'AI calling rate', it depends on which telephony provider, which LLM, and which voice you pick.

This is genuinely different from how most people think about calling cost, because a traditional phone bill really is one number per minute. An AI-handled call has that same telephony line, plus three more running in parallel behind it, each metered differently, some by minutes, some by tokens, some by characters of generated speech.

Why Convo360 doesn't quote a per-minute rate

Convo360 is a BYOK (Bring Your Own Keys) platform, you connect your own OpenAI, Anthropic Claude or Google Gemini key for the LLM, your own voice provider (Sarvam, Smallest.ai, ElevenLabs or OpenAI) for TTS, and your own telephony account for the call itself. Convo360 adds zero markup on messages, minutes or AI tokens, usage is paid directly to each provider at their own published rates.

This is a meaningfully different model from a closed platform that bundles all four components into one per-minute rate and adds its own margin on top. With BYOK, whatever your telephony, STT, LLM and TTS providers actually charge is what you pay, nothing added, but also nothing hidden or averaged away.

It also means the honest answer to 'what does an AI call cost' is 'it depends on your choices', not a fixed figure a vendor can print on a pricing page. Two businesses running the same Convo360 plan can have very different call costs simply because one picked a lighter LLM and a standard voice, while the other picked a heavier model and a cloned premium voice.

What actually moves the cost up or down

Since the total is a sum of four variable components, here's what changes each one in practice:

Component	What increases it	What reduces it
Telephony	Long outbound calls, international numbers	Shorter calls; check your provider's incoming and outgoing rates
STT	Long calls, multiple language switches	Concise caller responses, well-scoped call goals
LLM	Long system prompts, large knowledge base context per turn, verbose replies	Tight prompts, shorter agent replies, simpler decision trees
TTS	Long spoken replies, premium cloned voices	Concise scripts, standard voice tiers

Choosing a lighter LLM model for simple calls (like appointment confirmations) versus a heavier one for complex sales conversations is one of the biggest levers, since LLM and TTS scale with how much the agent actually says.

A worked example, without quoting rates

Picture two calls handled by the same AI receptionist for a clinic in Pune. The first is a 40-second call confirming tomorrow's appointment, short prompt, short reply, one language. The second is a 4-minute call where a caller asks several questions about treatment options, pricing, and directions to the clinic.

The second call costs meaningfully more across all four components: more STT minutes, more LLM tokens processed (both from the caller's questions and the longer replies), more TTS audio generated, and more telephony minutes. Neither call has a fixed 'per-minute' price, the total is a function of how much happened on the call, which is exactly why a blended, illustrative view of the four-part stack is more useful than chasing a single vendor number.

Now picture the same clinic running a hundred confirmation calls a week alongside a handful of longer enquiry calls. Most of the monthly total will come from the confirmation calls simply because there are more of them, even though each individual enquiry call costs more. This is why looking at total monthly usage across all four components, rather than a single per-minute figure, gives a much more useful picture of what a voice program actually costs to run.

Practical ways to reduce cost per call

Keep scripts tight. Write voice agent prompts to get to the point, shorter replies reduce both LLM tokens and TTS duration.
Use DTMF menus for simple routing. A DTMF menu block resolves in seconds compared to an open-ended conversation for basic call routing, cutting all four cost components at once.
Match the model to the task. A simple confirmation call doesn't need the same LLM as a complex sales conversation, pick per agent.
Compare incoming and outgoing rates. Incoming call charges depend on your telephony provider and plan, so check both rates on the provider's current price list before deciding whether callers should ring you or you should ring them.
Review Flow Logs and call transcripts for calls that ran unusually long, and tighten the flow at that point.

See the voice docs for the Call behaviour settings that also affect call length, like acknowledgement style and interruption handling.

Step by step

List your four providersIdentify your chosen telephony, STT, LLM and TTS providers — cost is the sum of what each one bills, not one number.
Match model weight to call complexityUse a lighter LLM for simple calls like confirmations and a stronger one only where the conversation needs it.
Tighten your scriptsShorter agent prompts and replies reduce LLM tokens and TTS duration together.
Compare inbound and outbound ratesCheck your telephony provider's current price list for incoming and outgoing charges, and design flows around the rates that apply to you.
Review long calls in Flow LogsRegularly check which calls ran long and adjust the flow to resolve common questions faster.

Example conversation

Why does one appointment call cost less to run than a sales call on the same number?

A confirmation call is short and uses a small part of the script, so it uses fewer LLM tokens, less speech-to-text and text-to-speech time, and fewer telephony minutes than a longer sales conversation with more back-and-forth.

See BYOK pricing on Convo360

Start free in minutes — no code, no credit card. Bring your own AI keys and pay no per-message markup.

Start 14-day free trial →See pricing

Frequently asked questions

Does Convo360 charge a per-minute rate for AI voice calls?

No. Convo360 is BYOK, so you connect your own telephony, LLM and voice provider accounts and pay each one directly at their own rates. Convo360 adds zero markup on minutes or tokens.

What makes up the cost of an AI voice call?

Four components: the telephony minute itself, speech-to-text to understand the caller, the LLM generating replies, and text-to-speech to voice the response. Each is billed separately by its own provider.

Are incoming AI calls free in India?

Not as a rule. Incoming call charges depend on your telephony provider and plan: Plivo's India price list shows inbound at ₹0.38 per minute, and Exotel says incoming calls are charged at your bill-plan rate. Check the provider's current price list. Convo360 adds no per-minute fee.

How can I reduce the cost of running AI voice agents?

Keep scripts concise, use DTMF menus for simple routing instead of open conversation, match a lighter LLM to simple call types, and review long calls in Flow Logs to find where scripts can be tightened.

Is a BYOK voice platform cheaper than a bundled per-minute platform?

It depends on your usage and provider choices — BYOK removes any platform markup, but the value comparison should be made against the actual rates your chosen providers charge, not a single headline number.