Voice AI Pricing Explained (and How to Cut Costs)
4 min read · Bolcho
Voice AI pricing confuses people because a 'minute' isn't one cost - it's the sum of several. Speech-to-text, the LLM, text-to-speech and telephony each meter separately, and most platforms add a per-minute markup on top, billed in dollars. Every pricing guide online quotes a headline rate ($0.07-0.50/min) but none show you where the money actually goes, and none price it for India. This one breaks a real minute down component by component in rupees, builds a line-by-line monthly bill, shows why Indian teams overpay 4-6x on USD platforms, and lists the levers that cut the bill.
What actually drives voice AI cost
A minute of conversation bills across four provider components, plus a platform fee. The big two are usually TTS (how many characters it speaks) and telephony (the actual phone line). The LLM - the part everyone worries about - is almost always the cheapest line.
The five cost components
- Speech-to-text (STT): per minute of audio transcribed.
- LLM: per token read (prompt + knowledge) and generated, every turn.
- Text-to-speech (TTS): per character synthesised.
- Telephony: per minute of the actual phone call.
- Platform fee: what many providers add per minute on top of the above.
The real per-minute breakdown (in ₹)
Here is where a rupee of a domestic Indian voice minute actually goes, on an efficient stack (Deepgram STT, a small fast LLM, Azure neural TTS, Plivo telephony). Prices move, so treat these as indicative - but the shape holds: TTS and telephony dominate, the LLM is ~8% of the minute.
Premium voices (e.g. ElevenLabs) push the TTS line 2-4x higher - which is why the voice you pick matters more to your bill than the model you pick.
~ ₹2.65/min
All-in raw cost, Azure voice
LLM = 8%
The model is the cheapest part
TTS + telephony = 62%
Where your bill really goes
Why Indian teams overpay on USD platforms
Global platforms bill in USD and add a per-minute platform fee - commonly $0.05-$0.13 - on top of provider costs. Converted to rupees and summed, a minute often lands around ₹8-14. A platform that meters near raw provider cost in ₹, with no per-minute markup and telephony included, lands 4-6x lower for the exact same conversation.
The markup is the difference
The underlying providers (Deepgram, OpenAI/Azure, ElevenLabs, carriers) are the same for everyone. The gap is the per-minute platform fee plus forex on dollar billing - which is exactly what near-cost ₹ billing removes.
A 5,000-minute month, line by line
Headline per-minute rates are abstract. Here is the same workload - 5,000 call-minutes a month - billed out in full, so you can see the real monthly number and where it differs. This is the reconciled bill most guides never show.
| Line item | Bolcho (₹ near cost) | Typical USD platform |
|---|---|---|
| STT | ~ ₹2,500 | bundled in rate |
| LLM | ~ ₹1,000 | bundled (+$0.02–0.05/min on some) |
| TTS | ~ ₹4,250 | bundled in rate |
| Telephony | ~ ₹4,000 | often billed separately |
| Platform fee | none | $0.05–0.13 / min |
| All-in per minute | ₹1.80–3.00 | ≈ ₹8–14 |
| Monthly total (5,000 min) | ~ ₹12,000 | ~ ₹40,000–70,000 |
| + 18% GST | ~ ₹14,160 | + forex + GST |
Same conversations, very different bill
At 5,000 minutes the gap is roughly ₹12k vs ₹40-70k a month. The conversations are identical - the difference is forex and the per-minute markup, and it compounds as you scale.
Subscription vs. pay-per-minute
Two billing shapes exist: usage (pay per minute) and subscription (a flat monthly fee, sometimes with included minutes). Neither is cheaper in the abstract - it depends on your volume and how predictable it is.
The rule of thumb: a flat plan wins when your usage is steady and above the plan's included minutes; pay-per-minute wins when volume is low, spiky, or you're still piloting.
- 1
Low or unpredictable volume
Pay-per-minute. You only pay for calls that happen, and a quiet month costs almost nothing.
- 2
Steady, growing volume
A subscription with included minutes can beat per-minute once you're reliably above the included bucket - just check the overage rate.
- 3
Watch the overage
Flat plans look cheap until you blow past the included minutes; overage rates (e.g. ₹2-4/min) are where the real cost hides.
How to estimate your monthly bill
A quick back-of-envelope: multiply expected call minutes per month by the all-in per-minute rate. 10,000 minutes at ₹2.40 is ~₹24,000; the same at ₹11 is ~₹1,10,000 - same conversations, very different bill.
- 1
Estimate minutes
Calls per month × average call length in minutes.
- 2
Pick a per-minute rate
Use the platform's all-in rate (provider + platform fee), converted to ₹.
- 3
Multiply, then buffer
Minutes × rate = your monthly voice spend. Add ~10-15% for retries and rounding.
- 4
Model growth
Re-run at 3× and 10× volume - the per-minute markup is what compounds.
How to cut your voice AI bill
The biggest levers, whichever platform you use:
- Cache repeated phrases (greetings, disclaimers, confirmations) instead of re-synthesising them - Bolcho does this automatically.
- Pick a standard neural voice over a premium one unless the brand truly needs it - it's the single biggest TTS lever.
- Route to the cheapest capable model per task instead of one premium model for everything.
- Keep prompts tight and use retrieval so the LLM reads fewer tokens each turn.
- Use good endpointing so you're not paying for dead air at the end of turns.
- Bill in ₹ near provider cost to remove the platform markup and forex entirely.
Pricing on Bolcho
Bolcho bills in rupees, near raw provider cost, with telephony included and no per-minute platform markup - the structure this whole guide argues for. You get the same providers everyone else uses, without the forex and the fee on top.
Estimate your own number with the breakdown above, then see it on a real call.
₹1.80–3.00/min
All-in, near provider cost
₹0 markup
No per-minute platform fee
Telephony included
Billed in ₹, no forex
Build it on Bolcho
Everything in this guide ships out of the box. Spin up an agent in minutes, free.
Frequently asked questions
How much does voice AI cost per minute?
It depends on the platform. USD platforms add a per-minute fee (often $0.05–$0.13) on top of STT/LLM/TTS/telephony, landing around ₹8–14/min all-in once converted. Bolcho bills in ₹ near raw provider cost, typically ₹1.80–3.00/min for a domestic Indian call.
What's the per-component breakdown of a voice minute?
On an efficient stack, indicatively: TTS ~₹0.85, telephony ~₹0.80, STT ~₹0.50, infra ~₹0.30, LLM ~₹0.20 - about ₹2.65/min raw. TTS and telephony are ~62% of the cost; the LLM is the cheapest part at ~8%.
Why is voice AI so much cheaper in ₹ than on USD platforms?
The underlying providers are the same. USD platforms add a per-minute markup and bill in dollars, so you pay forex on top. Near-cost rupee billing removes both, which is why the same conversation can cost 4-6x less.
Does the per-minute price include telephony?
On Bolcho, yes. On many USD platforms telephony is billed separately on top of the headline rate - always ask, because it's a big line (~₹0.80/min domestic India).
Is a subscription cheaper than pay-per-minute?
Only if your volume is steady and above the plan's included minutes. For low or spiky volume, pay-per-minute is cheaper because you only pay for calls that happen. Always check the overage rate on flat plans.
What are the hidden costs?
Premium/cloned voices (2-4x the TTS line), monthly phone-number rental, billing for failed or retried calls, forex and card fees on USD billing, concurrency caps that force a higher tier, and per-call rounding. Near-cost ₹ billing with telephony included avoids most of these.
How much does a premium voice (like ElevenLabs) add?
Roughly 2-4x the TTS line, so it can add ₹1-3/min on top. Worth it for a signature brand voice; for most agents a standard neural voice sounds natural and keeps the bill down.
What's the cheapest way to run voice AI in India?
Bill in rupees near provider cost with no per-minute markup, use a standard neural voice, cache repeated audio, route models per task, and keep prompts tight. That combination is what Bolcho is built around.