When evaluating voice AI platforms, the advertised per-minute rate is almost never the actual cost. Understanding what's includedand what's stacked on topis the only way to make an accurate comparison.
The Four Cost Components of Voice AI
Every voice AI call has four cost layers:
- STT (Speech-to-Text): Transcribing what the caller says. Typically $0.006–0.024/min on global providers.
- LLM (Language Model): Understanding and generating responses. Typically $0.002–0.04/min depending on model.
- TTS (Text-to-Speech): Generating the AI's voice response. Typically $0.015–0.03/min.
- Telephony: The phone call itself. In India, ₹0.50–1.50/min on most cloud telephony platforms.
Stacked individually at mid-market rates, these four components total ₹8–20 per minute before any platform margin.
What "Bundled" Actually Means
Platforms that advertise bundled all-in pricing have negotiated bulk rates across these providers and absorbed them into a single line. This is generally better valuebut verify what's actually included.
Questions to ask any vendor:
- Is telephony (the phone call) included in your per-minute rate?
- Is the LLM included, or do I need my own OpenAI/Anthropic key?
- Is Indian language STT included, or just English?
- Are there any per-call setup fees?
Agni pricing: ₹2–₹2/min is all-in. Voice (STT + TTS), LLM inference, emotion engine, and telephony are all included. No separate API keys required. No stacking.
The Hidden Cost: Latency-Related Failure
A call that takes 800ms to respond doesn't just feel badit has measurably lower completion rates. In India, where customers are often on variable mobile connections, high latency correlates directly with call abandonment. A cheaper platform that has worse latency may actually cost more in terms of outcome-per-rupee-spent.
Platform Fee vs Usage Fee
Most voice AI platforms have two cost components: a monthly platform fee and a usage fee. The platform fee is the cost of access; the usage fee is the cost of calls made.
At low volumes, the platform fee dominates the economics. At high volumes, the usage fee dominates. Calculate your break-even point before committing to any plan.
Example: At 2,000 calls/month, ₹2/min platform overhead from a ₹2,000 platform fee matters. At 20,000 calls/month, it's ₹0.20/minnoise.
The True Cost Comparison
| Platform | Advertised | All-in estimate |
|---|---|---|
| Agni | ₹2–₹2/min | ₹2–₹2/min (bundled) |
| Bolna | Variable | ₹7–12/min (estimated) |
| Retell (US-based) | $0.08/min | ₹18–30/min (+ telephony + LLM) |
| Vapi (US-based) | $0.05/min | ₹15–28/min (stacked) |
| Human agent (India) | ₹25K/mo salary | ₹18–24/min (fully loaded) |
At 10,000 calls per month at 3 minutes average duration, the difference between ₹6/min (Agni) and ₹25/min (US platform + stacking) is ₹5.7 lakh per month — ₹68 lakh per year.
