You must pick a vendor and a pricing model before the pilot, without knowing what a finished call will cost. AI voice agent pricing depends on four things: how the vendor bills, how long calls run, how many dials it takes to reach a person, and what sits outside the headline rate. Below are the models, a worked example and a hidden-fee list.
Five ways vendors charge for voice agents
Most vendors fit one of five models. Match the model to your call volume before you compare rates.
| Model | How it bills | Who it suits | Where it surprises you |
|---|---|---|---|
| All-in per minute | One rate for every connected minute | Small pilots, quick starts | Long calls and heavy redialing multiply the bill |
| Platform fee plus usage | A platform rate per minute, with phone, speech and model costs added on top | Teams that want control of each layer | The headline rate is only one layer |
| Build your own stack | You pay each provider directly and wire them together | Engineering-led teams at high volume | Engineering time and monitoring are not on any invoice |
| Seat or subscription | A monthly fee with minutes included | Steady, predictable volume | Overage rates once the bundle runs out |
| Minimum commit or outcome-based | A volume floor, or a price per result | Large, stable programs | You pay for the floor even in a quiet month |
Vendor rates and plan names change often. Where a vendor's rate is not published on its pricing page, ask for it in writing before you shortlist it.
Per-minute billing grows in a straight line. Cost equals billed minutes times the rate. If your campaign doubles its contacts, the bill doubles. If calls run twice as long, it doubles again. A collections lead at a 200-agent lender should price both numbers, volume and length, before choosing a model.
Worked example: a 10,000-contact campaign
This example is illustrative. Swap in your own numbers and your vendors' quotes. A collections lead at a 200-agent lender wants to remind 10,000 borrowers about upcoming payments.
Assumptions
- Each contact gets up to 3 attempts. Contacts who pick up early stop being dialed, so the campaign averages 2.2 dials per contact, which comes to 22,000 dials.
- 40% of contacts, or 4,000, are reached by a person. The other 18,000 dials go unanswered.
- An answered call averages 2.5 connected minutes. Billing rounds up, so it bills as 3.
- Unanswered dials bill one minute of phone line only.
- 15% of answered calls, or 600, go to a human for 4 minutes.
- 25% of answered calls end in a payment promise, which gives 1,000 outcomes.
- For comparison, a human agent spends 6 minutes on one answered call.
The arithmetic
- Billed AI minutes: 4,000 × 3 = 12,000.
- AI cost: 12,000 × the all-in rate per minute.
- Dead-dial cost: 18,000 × the phone-line rate.
- Handoff cost: 600 calls × 4 minutes × the human wage per minute.
For the wage, this example uses the US median of $20.59 per hour for customer service representatives in May 2024. That is about $0.34 a minute, or $816 for the handoffs. It is wages only. Wages vary a lot by region, so use your own rate.
| Scenario | All-in AI rate per minute | Phone rate per dead dial | Total | Per answered call | Per payment promise |
|---|---|---|---|---|---|
| Low | $0.10 | $0.01 | $2,196 | $0.55 | $2.20 |
| Mid | $0.15 | $0.015 | $2,886 | $0.72 | $2.89 |
| High | $0.25 | $0.02 | $4,176 | $1.04 | $4.18 |
The rates are placeholders chosen for a low, mid and high stack. They are not market prices, so replace them with real quotes. The high case costs 90% more than the low case at the same volume. Handoffs stay fixed at $816, so they are 28% of the mid-case total.
Human comparison. A human spending 6 minutes on one answered call costs about $2.06 in wages. With the placeholder rates, the AI call costs $0.55 to $1.04, handoffs included. Treat that gap as a prompt to test, since it rests on assumed rates.
Why cost per outcome matters. Cost per call is easy to misread. Gartner found that only a quarter of 432 AI customer-service use cases show positive ROI, and 42% cannot be measured. Spending is rising anyway: Gartner reports a 38% rise in customer-service AI spend against a 2% rise in overall budgets. Track cost per payment promise from day one, so you can show the number your finance team will ask for.
Hidden fees to watch for
The headline rate covers a connected minute. These ten items sit around it, each with what it is and what to ask. Twilio's figures below are its US prices.
- Dead-dial time. Ringing, voicemail and unanswered dials can bill as line time. Twilio, for example, lists US outbound toll-free calls at $0.0220 per minute and adds $0.0075 per call for answering machine detection. Ask when the clock starts and whether voicemail bills at the full AI rate.
- Rounding. A 61-second call can bill as two minutes. Ask whether billing is per second or per minute.
- AI-only silence. A call where no human spoke may still count as connected. Ask how "connected" is defined.
- Numbers and routing. Twilio lists US local numbers at $1.15 a month and toll-free at $2.15, and says charges vary by number type and country. Ask for rental fees and any international or local-number surcharge.
- Concurrency caps. Vendors limit simultaneous calls and charge to raise the limit. Ask for the cap and the overage price.
- Premium voices and languages. A cloned voice or a less common language can carry a higher rate. Ask for the rate in your language.
- Transfers. A warm handoff can bill both the AI leg and the human leg. Ask whether the AI stays on the line.
- Integrations. CRM sync and API access can sit on a higher plan. Ask which tier includes yours.
- Commits, setup and testing. Look for overage on committed volume, onboarding fees, custom-voice fees and test minutes billed before go-live. Ask for all four in writing.
- Recording and storage. Twilio charges $0.0025 per minute to record and $0.0005 per minute each month to store. Ask about retention length.
Compliance also adds cost, and the rules differ by region. In the US, the FCC ruled in 2024 that AI-generated voices fall under the TCPA, so outbound AI calls need consent. In the EU, Article 50 says people must be told when they interact with an AI system, including by automated phone systems. Ask which consent, do-not-call and disclosure tools are included and which cost extra.
Where Insight7 fits in the price picture
Insight7 Voice is a lower-cost voice engine for campaigns, now in beta. It uses the same ElevenLabs voices. Some controls also guard your spend. Calls where only the AI spoke are not counted as reached, and a carrier's temporary refusal does not use up a retry. Campaign setup includes calling hours, retry rules, consent marking and a test call, so confirm the current controls in the app.
Get a quote you can compare
Send every vendor the same pilot brief, so the quotes line up. Include:
- Contacts and calling hours
- Redial rules
- Share of calls handed to a human
- Languages
Ask each vendor for a per-minute rate, a price per answered call and a price per outcome.
Our advice is to ask for a price range early and put it in writing. Buyers use it to compare vendors and brief their own teams. Ask for a small paid pilot too. Banks will not pay just to learn whether it works, so tie payment to a measured lift.
An edtech buyer wanted prices before any pilot. Per-minute billing scaled in ways they could not predict, and they did not want to switch tools later.
Related reading: AI Voice Agent Implementation Timeline With a Week-by-Week Plan From Demo to Launch, What Is an AI Call Center Agent and How It Handles Inbound and Outbound Calls and How to Write an RFP for AI Call Center Software With a Free Template.
Questions people also ask
How much does an AI voice agent cost per minute?
Published rates vary by vendor and by which layers they include. The worked example uses placeholders of $0.10 to $0.25 per connected minute all-in. Those are not market prices. Ask each vendor for the all-in figure, then add dead-dial time.
How do I read a platform's pricing page?
Open the page and list each layer: platform, telephony, speech-to-text, language model and voice. Mark each as published, quoted or missing. Add the published layers per minute, then request the missing ones in writing. That sum is your real all-in rate, not the headline.
How much does ElevenLabs voice agent pricing run?
ElevenLabs publishes agent pricing on its own pricing page, and rates differ by plan and over time. Check the current page. Note whether the language model and phone line bill separately, because they raise the total. Ask for the all-in figure for your plan.
How much does OpenAI voice agent pricing run?
OpenAI bills audio by tokens, not minutes, so a per-minute figure depends on your call mix. Check OpenAI's current pricing page, then convert token rates to minutes using a test call. Phone lines are billed separately by your telephony provider.
Is an AI voice agent cheaper than a human call agent?
It can be, but test it. In the worked example, placeholder rates gave an AI call cost of $0.55 to $1.04 with handoffs included, against about $2.06 in human wages alone. Your real rates decide the result, so compare cost per outcome from a pilot.
What is the difference between per-minute and per-call pricing?
Per-minute pricing charges for connected time, so cost grows with call length. Per-call pricing charges one fee, so length stops mattering. It suits short, scripted calls like reminders. Ask what counts as a call, and whether a voicemail or a two-second pickup does.
Are AI voice agents billed for unanswered calls and voicemail?
Sometimes. Telephony vendors can bill per minute on calls, and some platforms apply the full AI rate once the line connects. Ask how "connected" is defined. Before you sign, make 50 test dials and compare the invoice to the call log.
What is the cheapest way to build an AI voice agent?
Building your own stack can cut the per-minute cost, because you pay each provider directly and skip a platform fee. Whether that pays off depends on volume. Engineering time and monitoring are real costs, so price a managed platform first and compare.
How much does an AI voice calling agent cost per month for a small team?
Multiply your expected connected minutes by a vendor's all-in rate. At 1,000 minutes, a placeholder rate of $0.10 to $0.25 gives $100 to $250. That is about 330 calls billed at 3 minutes each. Add number rental, dead dials and any plan minimum.
How do I estimate the cost of an outbound voice AI campaign before a pilot?
Count answered calls, not contacts. Multiply contacts by your expected answer rate, then by billed minutes rounded up, then by the all-in rate. Add dead dials times the line rate, plus handoff minutes. Run a pilot of a few hundred contacts to replace guesses with real figures.
Price the outcome you pay for
The sticker rate covers only the connected minute. Your real cost depends on how many dials it takes to reach a person, how long calls run, and the fees around the call. Put every vendor's quote on the same pilot brief and divide by results, not minutes. Rule of thumb: price the outcome you pay for, never the minute.
