AI Voice Agent Business: Revenue Cases and the Model Behind Them

Nine businesses in our index get paid because a machine holds a phone conversation, and only three publish a company-level run rate: $120K ARR, $1M ARR and $125M/yr. The per-client rate card is firmer than the revenue — $500–1,500 per business per month, or $250 per seat — and that is where the model works or quietly dies. Nine cases is a thin file: read the grade before the dollars.

Contents

• ·How did we pick these nine AI voice agent businesses?
• ·What does an AI voice agent business actually earn?
• ·How do you price one: per business, per seat, or per outcome?
• ·Why do most voice agent businesses stall before the second invoice?
• ·Where does this model break?
• ·Who should start one, and who should not?
• ·Frequently asked questions

How did we pick these nine AI voice agent businesses?

One rule: somebody pays because a machine talks on a phone line. Our two voice sets hold 34 unique records, and 25 are faceless channels, dubbed podcasts, audiobooks and avatar accounts where synthesis is only a production cost. Those are a tools story. Nine are businesses where the call is the product.

Grades: ✅ third-party verified, 🗣 founder-reported, 📎 creator-relayed, 🔮 unproven. The nine split 2 ✅, 3 🗣, 4 📎 and none unproven — respectable for filmed teardowns, and still nine. Two carry an asterisk: ElevenLabs is the supplier, not an operator, and the AI chat agent is a control case.

A tools page asks which voice sounds best; this one asks who signs a recurring invoice, and for how long. More teardowns sit in our AI agencies hub.

What does an AI voice agent business actually earn?

From $120K ARR to $125M/yr, with very little in between that publishes a monthly figure. Three of the nine give a company-level run rate, and their median is Clarvo's $1M ARR — a three-point median is not a market, so treat it as one data point wearing a decimal place.

Do not average that column. ElevenLabs sells picks and shovels to everyone else in it, and "some still in pilot" is doing quiet work in GigaML's row. Sanas took $30M at a $150M post while pre-product, then grew to roughly 200 people in India to serve customers worth about $500K each. Plan against Setter AI's numbers instead: $10K MRR from 40 customers at a cost base under 10% of revenue, earned by wrapping a YC company's voice API rather than building a stack.

How do you price one: per business, per seat, or per outcome?

Four shapes appear here. Flat per business ($500–1,500 a month), per seat (Clarvo's $250, where a 100-seat deal is $300K ARR), per engagement (that freight-dispatch receptionist at six figures over six months), and per booked outcome (Infinity Pilot's roughly $3,000 average deal).

The flat retainer is where nearly everyone starts, and its arithmetic is instructive. Setter AI's $10K MRR across 40 customers averages about $250 each per month — our division of their own two figures — below the bottom of Dan's $500–1,500 range. That gap is the distinction: the range prices a service, configured and babysat per business.

Per-minute is the wrong axis for the operator, even though vendors bill that way. Nobody here publishes a per-minute rate, and the only margin figure in the file is Setter AI's sub-10% cost base: at $500–1,500 a month, minutes are a rounding error and onboarding is the real cost. Check the vendor rate card, then price against the client's missed calls. Per-resolution billing is fashionable in enterprise deals; no case in our data reports a dollar from it.

Why do most voice agent businesses stall before the second invoice?

Building was never the constraint. Dan, who rates the voice receptionist A-tier, puts it bluntly: everyone is busy building voice agents and almost nobody knows how to sell one. Every case here with money attached solved distribution first and treated the agent as delivery.

Setter AI sold the first deal for $500 before there was a dashboard to log into. Infinity Pilot AI turned 227 leads into 69 appointments and 27 deals at a 40% close rate, $89,000 contracted in 75 days — the agent's job there is speed, while trained humans close. GigaML won Fortune 500 accounts with 4–5 engineers because an internal agent kills its onboarding boilerplate, not because its voice is better. Enterprise voice revenue is headcount you have not hired yet.

Where does this model break?

Three places: pilots that never convert, a commodity ceiling, and the phone's own rulebook. We file GigaML as a ⚠️ cautionary tale despite the logos precisely because pilots can stay pilots for a year, and a Fortune 500 name in a deck is not a paid invoice.

The ceiling is documented. The consultant behind the 97% audit tracked 1,000+ vibe-coded apps and found 97% dead, abandoned, breached or under $500/mo after six months, while the survivors he works with are dull: $1,400 a month per shop for HVAC routing, plus that six-figure freight-dispatch receptionist. His test — if you can describe your ICP in one tweet, so can 10,000 other people — is the sharpest churn warning in the set, and "AI receptionist for local businesses" is a tweet.

The AI chat agent is the control: same buyer, lower barrier, Intercom already there, no revenue figure from anyone. A low barrier helps you sell on Monday and hurts you at renewal.

Then the rulebook. The FCC's declaratory ruling of 8 February 2024 made AI-generated voices in calls "artificial" under the TCPA, so telemarketing robocalls need prior express written consent (FCC announcement). An inbound receptionist barely notices; an outbound setter dialing bought lists is the exact shape that ruling targets.

Who should start one, and who should not?

Start if you already reach businesses that lose real money on missed calls, and if you can sell before you build. Do not start if the plan is a demo video plus a cold DM list, or if your differentiator is being the cheapest AI receptionist in a town that already has four.

Our read, in order: pick a vertical specific enough that a tradesperson recognises themselves in it, charge per business rather than per minute, and let the agent book while a human closes. The honest gap is retention — none of the nine publishes churn, so $500–1,500 a month is a price we can evidence and a lifetime we cannot.

On that second half, AI agency revenue collects client pricing that survived a renewal, AI agent for business covers what small companies buy, and AI agents that make money grades the wider field. More in our AI agencies hub.

Frequently asked questions

How much does an AI voice agent business make?

Three of our nine cases publish a company-level run rate: $120K ARR from 40 customers, $1M ARR from a seat-priced dialer, and $125M a year from the vendor. The service tier reports a price rather than a profit: $500 to $1,500 per business per month. Plan against the documented small end.

What should I charge for an AI receptionist?

The only rate card in our data is $500–1,500 per client per month, relayed by a creator rather than audited. Seat pricing appears once, at $250 per seat for an outbound dialer. Charge per business rather than per minute: the one margin figure available shows costs under 10% of revenue, so minutes are not what constrains your price.

Is an AI voice agent business still worth starting?

Yes for a narrow vertical with expensive missed calls, no for a generic local-business offer. The documented survivors in our set are unglamorous — freight dispatch, HVAC routing — and a low build barrier invites thousands of identical competitors. The pain and the willingness to pay are real; differentiation comes from the niche.

Voice agent or chat agent, which sells better?

On this evidence, voice. The voice receptionist has a published price range and an A-tier rating from the creator who covers both models; the chat agent has neither and faces established incumbents. The suggested combination is chat to qualify and voice to close, though no case in our file reports revenue from that stack.