Voice AI economics / field note

One minute. Five meters.

Q: How do you calculate voice AI cost per minute?

Add the per-minute platform, telephony, speech-to-text, text-to-speech, and language-model rates. Keep fixed subscriptions separate, because they change effective cost per minute as usage changes.

Q: Is cost per minute enough to compare voice AI systems?

No. Cost per minute describes usage, not successful outcomes. Include fixed fees, the share of calls resolved correctly, and human work created by unresolved calls.

Q: Which voice AI costs are usually missed?

Teams often miss phone-number fees, minimum commitments, silence and hold time, failed calls, retries, call transfers, post-call work, monitoring, and human escalation.

Q: How do you convert voice AI cost per minute into cost per resolved call?

Add monthly voice AI spend and human handoff cost, then divide by the number of calls the AI resolved correctly without a person.

Voice AI cost per minute is the sum of platform, transport, transcription, synthesis, and model usage. The useful business metric starts one level deeper: loaded cost per correctly resolved call.

Open the full calculator →

The complete formula

Move from a rate card to an outcome.

1. Stack cost / minute
platform + telephony + STT + TTS + LLM

2. Monthly AI spend
(calls × minutes × stack rate) + fixed fees

3. Human handoff cost
unresolved calls × handoff minutes × hourly cost ÷ 60

4. Loaded cost / resolution
(AI spend + handoffs) ÷ AI-resolved calls

Illustrative sensitivity test

The cheapest minute does not always win.

Each scenario holds volume and outcomes constant at 5,000 four-minute calls, 70% resolution, $200 fixed fees, and six minutes of $30/hour human work for every unresolved call. Only the example stack rate changes.

Illustrative lean stack $0.060/min $1.69 loaded / resolution → Illustrative base stack $0.105/min $1.94 loaded / resolution → Illustrative robust stack $0.200/min $2.49 loaded / resolution →

Decision rules

Separate fixed and variable.
A subscription changes effective per-minute cost at every volume.

Measure real resolution.
Transfers, repeats, and incorrect completion are not automated outcomes.

Count cleanup.
Human escalation belongs in the operating cost, not in a footnote.

Build the input sheet

Collect evidence before comparing stacks.

Export connected minutes and call outcomes for one narrow workflow. Match every provider charge to the same month, region, transport, and plan. Then price unresolved calls from observed handling time.

Calculate the loaded outcome cost → Use the AI receptionist worksheet → Test the same stack's latency →

How do you calculate voice AI cost per minute?

Add the per-minute platform, telephony, STT, TTS, and LLM rates. Keep subscriptions and other fixed fees separate so volume does not hide them.

Is cost per minute enough to compare voice AI systems?

No. A usage rate ignores fixed fees, resolution quality, transfers, repeat contacts, and human cleanup. Compare loaded cost per correctly resolved call.

Which voice AI costs are usually missed?

Include phone numbers, minimum commitments, silence and hold time, retries, failed calls, monitoring, transfers, post-call work, and human escalation.

How do you convert voice AI cost per minute into cost per resolved call?

Add monthly AI spend and human handoff cost, then divide by calls the AI resolved correctly without a person.