AI voice agent cost: per-minute rates, build price and what a call really costs

How much does an AI voice agent cost in total?

The total is a usage bill that grows with call minutes, a small set of fixed monthly charges, and a build cost you pay once. Most pricing pages show only the first layer, which is why budgets based on a headline rate tend to come in low.

The usage layer is easy to estimate once you know your call volume and average call length. The fixed layer is small for most companies but jumps when you outgrow a plan's concurrency limit or daily call cap. The build layer varies most, because it depends on how many systems the agent reads from and writes to, how many call types it handles and how strict the compliance requirements are.

What does an AI voice agent cost per minute on each platform?

Published rates in October 2026 cluster around ten to fifteen cents per minute once every component is counted, but vendors package the components differently. The table lists figures we read on each vendor's own pricing page this month. They are third-party prices, not ours, and they change often.

Retell AI is the clearest example of a stacked price. Its page lists $0.055 per minute for its voice infrastructure, then adds the model (for example $0.025 per minute for a small, fast model or $0.064 for a mid-tier OpenAI or Claude model), platform voices at $0.015 and US telephony at $0.015. Its own calculator shows about $0.11 per minute for a basic configuration. Choosing a premium voice or a larger model can push the rate well above that.

Vapi works the same way with a different split: $0.05 per minute for hosting, with transcription, model, voice and telephony passed through at cost. Using the example component prices on its page (Deepgram transcription at about $0.01, OpenAI models from about $0.008 to $0.045, ElevenLabs voices from about $0.015 to $0.024) plus Twilio's outbound rate, a call works out to roughly $0.10 to $0.14 per minute depending on the model and voice.

Why is cost per minute not the same as cost per call?

For outbound work, you pay for every attempt that connects to something, not only for the conversations that finish the job. A campaign that confirms appointments or follows up on quotes reaches voicemail, wrong numbers and people who hang up in the first few seconds, and each of those consumes billed minutes.

The number that matters is cost per completed outcome: confirmed appointments, booked meetings, finished surveys. To get it, divide the monthly usage bill by the number of calls that produced the outcome you wanted. Two agents with the same per-minute rate can differ widely here, because one keeps voicemails short and stops retrying after a sensible limit while the other does not.

Fixed limits matter as much as rates. Bland AI's Start plan caps usage at 100 calls a day and 10 concurrent calls, so a list of 4,000 dials a month (about 180 per weekday) does not fit on it even though the per-minute rate looks fine. Retell AI includes 20 concurrent calls and charges $8 per extra line per month, and Vapi charges $10 per extra line. If your call window is short, for example confirmations between 9 and 11 in the morning, concurrency rather than minutes sets your plan.

• Voicemail minutes: decide per campaign whether to leave a message and keep it under 20 seconds.
• Retries: cap attempts per contact and space them out, so unreachable numbers stop costing money.
• Transfer minutes: some platforms bill the time a call is bridged to your staff separately.
• Daily caps and concurrency: size the plan to your busiest calling hour, not your monthly total.
• Recording and storage: carrier and platform recording fees are small per minute but add up at volume.

Worked example: what does an outbound confirmation campaign cost per month?

Here is a monthly budget for a clinic group or service business confirming appointments by phone. The call outcomes are assumptions chosen to show the math, not benchmarks, so replace them with your own call reports. The rates come from the vendor pricing pages in the table above.

Assume 4,000 dials a month. Half reach a person and last about 1.5 minutes (3,000 minutes), 30% reach voicemail and leave a 30-second message (600 minutes), and 20% go unanswered with no billed time. That is 3,600 billed minutes and 2,000 conversations, of which the confirmations, reschedules and cancellations are the outcomes you want.

At Retell AI's calculator rate of about $0.11 per minute, usage comes to about $396, plus $6 for three numbers, so roughly $402 a month or about $0.20 per conversation. On Vapi, the same minutes at $0.10 to $0.14 come to about $345 to $514 before plan fees. On Bland AI, the volume exceeds the Start plan's 100 calls a day, so the Build plan applies: 3,600 minutes at $0.12 is $432 plus the $299 platform fee, about $731 a month.

What does AI voice agent development cost?

Published build ranges run from a few thousand dollars for a single-purpose agent to six figures for an enterprise voice platform. The spread comes from scope: how many call types, how many systems the agent writes to, how many languages and how much compliance work is involved.

Groovy Web's July 2026 guide gives the lower ranges. RaftLabs' guide (updated June 2026) is more expensive at each tier and adds timelines, which is useful for planning. RaftLabs prices custom voice systems, while Groovy Web's lower tiers fit builds on an existing voice platform such as Retell AI or Vapi, where the call handling already exists and the engineering goes into integration and rules. Groovy Web lists integration depth among the main cost drivers, and that matches our own scoping: writing outcomes reliably into HubSpot, Salesforce or a practice management system takes more engineering than the conversation itself.

What does an AI voice agent cost in the first year?

Add the build to twelve months of running costs and a maintenance allowance. Using the worked example above: a build at $20,000 (inside Groovy Web's multi-intent band) plus $402 to $731 a month in usage and plan fees gives roughly $24,800 to $28,800 for year one, before any maintenance retainer. RaftLabs puts ongoing maintenance at about $500 to $2,000 a month, so budget a line for prompt updates, new call types and fixes even if you handle some of it in-house.

Year two is usually cheaper because the build is paid for, which is why the comparison against staff time should be made over at least two years. Measure the baseline first: how many calls staff place today, how long each takes including dialing and notes, and how many slots or deals are lost when calls are not made.

Which compliance costs change the total?

Healthcare calls and outbound consent rules are the two items that move a budget most. If calls involve protected health information, every vendor that handles it needs a signed Business Associate Agreement. Vapi sells HIPAA compliance as a $2,000 per month add-on, Bland AI offers a BAA on its Enterprise plan, and Retell AI lists custom BAA terms under Enterprise, so the cheapest self-serve plans are usually off the table for clinical calls.

For outbound calls, the FCC ruled on February 8, 2024 that AI-generated voices count as artificial voices under the TCPA, so those calls need the called party's prior express consent. That turns consent into an engineering cost: a consent field the dialer checks before every call, a suppression list for opt-outs, and logs of each dial and outcome. Some states also require telling people they are speaking with an AI, and two-party consent states apply to recorded calls. This is not legal advice; confirm your call types with counsel.

• BAA-eligible tiers for the voice platform and the model provider on any call with health information.
• A pre-dial consent check and an opt-out suppression list for every outbound campaign.
• An AI disclosure and recording notice at the start of each call.
• Retention rules for recordings and transcripts, with storage priced in.

How do you lower the running cost without hurting the calls?

Most savings come from shorter calls and fewer wasted attempts, not from switching to the cheapest platform. These are the controls we design into outbound agents from the start.

• Model routing: a small, fast model for confirmations and simple intents; a larger one only where the call needs reasoning.
• Tight scripts: one job per campaign, short turns and a fixed set of outcomes, so calls end when the job is done.
• Text first: send a confirmation SMS and only call the people who do not reply, which removes many calls entirely.
• Attempt limits and spacing enforced by dialer rules outside the model.
• A cost cap per call that ends a stuck conversation and creates a task for staff.
• Weekly review of cost per completed outcome by campaign, not just total minutes.

When is an AI voice agent not worth the cost?

It is usually not worth building one for a few dozen calls a week, for calls that are mostly negotiation or complaints, or where you cannot show recorded consent for the people on the list. In the first case staff time is cheaper than a build; in the second the agent would hand most calls to a person anyway; in the third the legal risk outweighs the savings.

If your need is mainly answering inbound calls, a packaged receptionist with a monthly plan may cost less than a custom agent. Our AI receptionist cost guide compares those plans. A custom voice agent earns its build cost when call volume is steady, outcomes are structured and the results need to land in your CRM or scheduling system without anyone retyping them.

How do we price an AI voice agent build?

We quote a fixed price after a free review of your call lists, consent records, phone setup and the systems the outcomes must reach. The review produces a call design for each campaign, a minutes estimate built from your own numbers and a plan sized for your busiest calling hour.

The voice platform, Twilio and model accounts are opened in your name, so usage is billed to you at vendor rates and you keep the scripts, configuration and call logs. A person handles disputes, payments and anything sensitive. For broader build and running cost ranges across other AI systems, see our AI implementation cost guide.