What it costs you, and where to price it.
Every number below comes from a real call on the live agent: 250 seconds, 50 turns, 1,895 characters spoken. Vendor prices read November 2026.
01 — Your cost floor
Measured, not estimated
The whole stack on one 4-minute conversation. There is no platform fee underneath it — the server is already paid for, so this is close to the true marginal cost of a call.
| Component | Rate | This call | Per minute |
|---|---|---|---|
| Deepgram Aura-2 (speech out) | $0.030 / 1k chars | $0.057 | $0.0137 |
| Deepgram Nova-3 (speech in) | $0.0077 / min | $0.032 | $0.0077 |
| Twilio inbound | $0.0085 / min | $0.035 | $0.0085 |
| DeepSeek (thinking) | $0.28 / $0.42 per M | $0.041 | $0.0098 |
| Phone number | $1.15 / month | — | — |
| Server share | fixed, already paid | — | — |
| Total | $0.165 | $0.0397 |
Speech out is the biggest line at 35%. Short answers cut more cost than any vendor switch. DeepSeek is approximate — the pipeline does not log tokens yet, and prompt caching will pull it lower.
02 — What the market already pays
Three anchors, wildly different
Your price has to sit between the do-it-yourself platforms and the human alternative, and be explainable against both.
| Option | What it costs | What the buyer gets |
|---|---|---|
| DIY platform (Retell, Vapi, Synthflow) | $29–$449/mo, or $0.07–0.36/min | Software. They write the prompt, wire the number, fix it themselves. |
| AI answering service | $0.15–0.35/min retail | A configured agent, usually generic, no custom intake. |
| Human answering service | $1.50–$3.60/min | 900 minutes ≈ $1,800/mo. Your real competitor. |
Anchor against the human, never the platform. A buyer comparing you to Retell is buying software; a buyer comparing you to their answering service is buying their phone answered.
03 — The offer
Build fee, then flat monthly with included minutes
Per-minute billing invites the platform comparison and punishes you for the client being busy. Included minutes with a priced overage keeps the invoice predictable and the margin intact.
| Tier | Setup | Monthly | Included | Overage | Your cost* | Margin |
|---|---|---|---|---|---|---|
| Starter — one number, lead capture, email | $2,500 | $299 | 300 min | $0.35 | $18 | 94% |
| Growth — custom intake, CRM/calendar, research | $3,500 | $599 | 900 min | $0.30 | $42 | 93% |
| Pro — multi-number, integrations, monthly tuning | $5,000 | $1,199 | 2,500 min | $0.25 | $106 | 91% |
*Cost at full included usage, including the number and a $5 server share. Margin is gross — before your own time.
04 — Run a client
What this account actually makes
05 — Rules that protect the margin
- Cap included minutes. One chatty client can eat an account. The cap is what makes the flat fee safe.
- Overage at 7–8× cost. Heavy users stay profitable instead of being subsidised.
- Scope support, don't promise it. Prompt tuning monthly, not on demand. Every weekly email costs more than the minutes.
- Don't sell minutes. Sell "every call answered and written down". Minutes are how you buy it, not what it is.
- Setup fee or commitment, never neither. Free builds attract clients who leave in month two.
- Never publish prices on the site. The quote is a conversation; the site qualifies the work.
07 — See it working
Films cut for arnowsolutions.com and social, playable here. The first is the two-minute piece: what a missed call actually costs a business, and what answering every one of them looks like. The second is the same story cropped for Reels, TikTok and Shorts. The rest are earlier cuts and one vertical of the shop film.
Half-width vertical for the social cut so it doesn't tower over the 16:9 films. Earlier cuts (v1–v4) and the shop-film vertical stay on the media host: media.srv1738752.hstgr.cloud/voice-agent-film/.
06 — What to do next
- Pick one vertical for the first three clients — the intake questions only get sharp inside one trade.
- Charge the setup fee on the first three even if discounted. Free pilots teach you nothing about willingness to pay.
- Track minutes from day one (the dashboard already stores seconds per call) so the second invoice is priced from data.
- Raise the Growth price at client five, not before. Early clients buy the story; later clients buy the proof.