AI cost and operations
Voice AI pricing: why the list price is not the bill
Vendors sell different units, so the cheapest rate on a pricing page is often not the cheapest bill.
How much does voice AI cost?
Voice AI is usually priced per minute of call time, per customer handled, or as a monthly plan with an allowance. Goodcall's AI answering service started from $79 a month in November 2026. The list price is rarely the whole bill: telephony, overage, integration and testing decide what you actually pay.
The short version
- Voice AI is priced in three main units: minutes of call time, customers handled, or a monthly plan with an allowance.
- Per-minute and per-customer rates cannot be compared until you know your call volume, unique callers and average call length.
- Goodcall's Starter plan covered 100 unique customers a month in November 2026, then charged 79¢ for each extra customer.
- A voice AI list price covers the platform, not the integration, testing and upkeep that make an agent reliable.
- The fairest way to compare voice AI with a receptionist is cost per resolved call, using full employment cost rather than the wage.
How much does voice AI cost?
Voice AI has no single market price, because vendors do not sell the same unit. Some charge for minutes of call time, some for customers handled, and some for a monthly plan with an allowance. A rate on a pricing page tells you little until you convert it to your own call pattern.
One published plan shows the shape. Goodcall advertises an AI answering service from $79 a month. Its Starter plan covers 100 unique customers a month, then charges 79¢ for each extra customer. That unit is customers, not minutes, which changes what drives the bill.
Retell AI publishes a list price on its own pricing page, and Twilio states its list price in the FAQ on its conversational AI pricing page. Read both directly, because list prices change and a copied number goes stale faster than the page it came from.
What are the main voice AI pricing models?
Three models cover most vendor pricing: per minute, per customer, and a monthly plan with an allowance. A custom build adds a fourth. Each one moves a different risk onto you, and the table below shows what you pay for and where each model catches buyers out.
How do voice AI pricing models compare?
| Model | What you pay for | Suits | Watch for |
|---|---|---|---|
| Per minute | Minutes of call time, sometimes with telephony as a separate line | Uneven volume and short calls | Long calls, hold time and a talkative agent all raise the bill |
| Per customer | Each unique customer handled in the month | Callers who ring back several times | Many one-off callers, and how wrong numbers or spam are counted |
| Monthly plan with allowance | A fixed fee, plus overage above the allowance | Steady, predictable volume | Unused allowance in quiet months, overage in busy ones |
| Custom build | Engineering time, then usage on each underlying service | Unusual call flows, or systems a platform cannot reach | Upkeep, which does not stop after launch |
What does a per-minute voice AI rate include?
It depends on the vendor, and the rate on the page may not cover the whole call. A voice agent is a chain of services: telephony carries the call, speech-to-text hears the caller, a language model decides the reply, and text-to-speech says it. Each link has a cost. Some vendors bundle all of them into one rate. Others list them separately, so the headline figure covers only part of the chain.
Ask for the all-in cost of one minute of a real call, with your region, number type and chosen voice. Then ask how the minute is measured. Rounding up, ring time, hold time and silence can all count as usage. A transferred call may keep the meter running after a person has picked up.
Your own design moves the cost too. An agent that reads out long answers uses more minutes than one that gets to the point. A slow reply keeps the caller on the line for longer. On platforms that pass through model costs, a long prompt sent on every turn adds to each minute.
How do you compare per-minute and per-customer pricing?
Convert both to a cost per resolved call. Minutes and customers are different units, so the rates cannot be compared directly. You need three numbers from your phone system: calls a month, unique callers a month and average call length.
Goodcall's published plan makes a clean worked example. Taking $79 as the Starter price with 100 customers included, 150 customers cost $118.50 and 300 customers cost $237. Above the allowance the whole plan works out at exactly 79¢ a customer, so the $79 behaves like a monthly minimum. At 50 customers it comes to $1.58 each. This is arithmetic on the plan figures, not a number Goodcall states.
A per-minute price has a different driver. The monthly cost is calls, multiplied by average length, multiplied by the all-in rate. Double the call length and that bill doubles, while a per-customer bill does not move unless the plan also limits minutes. If customers ring back three times about one job, a per-customer unit favours you. If most callers ring once, that advantage goes.
How do you work out your own voice AI cost?
- Export one typical month from your phone system, plus one busy month if your trade is seasonal.
- Count calls, unique callers and average call length.
- Estimate how many calls the agent could finish alone, and check that estimate against real calls.
- Apply each vendor's unit to those numbers, with telephony and overage included.
- Add the monthly cost of whoever handles escalations and maintains the agent.
- Divide the total by resolved calls. That is the figure to compare.
What costs sit outside the voice AI list price?
The list price covers the platform, not the work of making it reliable on your calls. The main costs outside it are integration, testing, escalation handling and upkeep. For a simple message-taking job these can be small. For an agent that books, looks things up or takes payment, they can exceed the platform fee.
Integration comes first. An agent that only takes messages needs little. One that books appointments needs your calendar, your rules and a way to handle the slot that vanished mid-call. Every system it touches is another thing to build, secure and keep working.
Testing is the cost buyers skip. A demo call in a quiet room shows very little. Real calls have accents, road noise, people talking over the agent and questions nobody scripted. You need to test on realistic calls, then test again whenever a prompt or model changes.
Then there is the cost of a wrong answer. An agent that misquotes a price or books the wrong slot does not appear on any invoice, but somebody pays to fix it. Count the staff time for escalations and failed calls, because it does not fall to zero.
Is voice AI cheaper than a receptionist?
It can be for routine calls at volume, but only if you compare full costs on both sides. Count the employment cost of the person, not just the wage, and count the staff time the AI still needs. The honest unit is cost per resolved call.
For an in-house receptionist in the US, start with two Bureau of Labor Statistics sources. Its Occupational Outlook Handbook gives the May 2025 national median for receptionists. Its Employer Costs for Employee Compensation survey, for June 2026, shows what employing someone costs beyond the wage. That second figure is an average for all private industry workers. It is not specific to receptionists or small firms, so treat the result as an estimate.
A virtual receptionist service is often the closer comparison for a small firm, because it is also billed by usage. Ruby, for example, sells a virtual receptionist plan of 100 minutes per month. Set that allowance against your own monthly call minutes before you look at the price.
Some differences never appear in a rate. A person at the front desk does other work and handles the call that fits no script. An AI agent answers out of hours and takes several calls at once. Decide which of those your business needs before comparing numbers.
When is voice AI not worth paying for?
It is not worth it when call volume is low, when calls are rarely routine, or when a wrong answer is expensive. If you take a handful of calls a day, voicemail with a prompt call-back or a virtual receptionist may cost less and carry less risk.
The same applies when most calls need judgement. Complaints, disputes and sensitive conversations are poor candidates, because the agent either escalates them, so you pay for both, or handles them badly. Look at the easy fix first. If callers mostly want opening hours, a booking link or an order status, a text message or a better website may remove the call altogether.
What should you ask a voice AI vendor before signing?
- What is the all-in cost of one minute, or one customer, including telephony?
- How is usage rounded, and are ring time, hold time and transfers billed?
- What happens at the allowance: a hard stop or automatic overage?
- How are spam calls and wrong numbers counted?
- Can we set a spending cap and a usage alert?
- Can we export recordings and transcripts if we leave?
- What notice do we get of a price change or a model change?
Questions people ask
Why is my voice AI bill higher than the advertised rate?
Usually because the advertised rate covers only part of the call. Telephony may be billed separately, minutes may round up, and hold time or transfers may count as usage. Overage above a plan allowance adds more. Ask the vendor for an itemised bill from a real pilot month, then compare it with the rate you were quoted before you commit to a longer term.
Is per-minute or per-customer pricing better?
Neither is better in general. Per-customer pricing favours businesses with long calls, or callers who ring back several times, because neither adds to the count. Per-minute pricing favours short, one-off calls and uneven volume. Take one month of your call log, apply both units, and compare the cost per resolved call. The answer depends on your calls, not on the vendor's rate.
Should we use a voice AI platform or build a custom agent?
Start with a platform unless you have a specific reason not to. A platform is faster to test and carries much of the upkeep. A custom build makes sense when calls depend on systems or rules a platform cannot reach, or when you need full control over data and models. It costs engineering time up front, and the maintenance does not stop after launch.
How do we stop a voice AI bill running away?
Set limits before launch. Ask for a spending cap or a usage alert, put a maximum length on each call, and make sure spam calls are blocked or ended quickly. Review the longest calls every week for the first month, because they show loops and confusion early. If a vendor cannot offer a cap or an alert, treat that as part of the price.
Do these prices apply outside the United States?
Treat them as a starting point only. The plan figures quoted here are in dollars and the Bureau of Labor Statistics data covers the US. Telephony rates differ by country and number type, and call recording rules differ too. If you are outside the US, ask each vendor for the all-in cost in your region, and use local wage data for any staffing comparison.
What should a pilot measure before we sign a longer contract?
Measure how many calls the agent finishes without help, the cost per resolved call, how often callers ask for a person, and what went wrong in the calls that failed. Run it on real traffic, not a test script, and for long enough to include a busy period. A pilot that reports only minutes handled tells you what you spent, not what you got.
Related on this site
Sources
- AI Phone Agent Pricing | Retell AI, Vendor's own published list price, read November 2026. · Retell AI, data from 2026
- Conversational AI solutions pricing | Twilio, Vendor's own published list price, stated in the FAQ on its pricing page, read November 2026. · Twilio, data from 2026
- Goodcall Pricing | AI Answering Service from $79 a Month, Vendor's own pricing page, read November 2026. Starter covers 100 unique customers a month, then 79¢ each. The price appears on a plan card, not in a sentence, so the quote is the plan-card text. · Goodcall, data from 2026
- Ruby Receptionist Pricing: Scalable Virtual Support, Vendor's own pricing page, read November 2026. This is the 100 minutes per month virtual receptionist plan. The price appears on a plan card, not in a sentence, so the quote is the figure alone. · Ruby, data from 2026
- Receptionists : Occupational Outlook Handbook: : U.S. Bureau of Labor Statistics, BLS Occupational Employment and Wage Statistics survey, May 2025 national median for the occupation. · U.S. Bureau of Labor Statistics, data from 2025
- Employer Costs for Employee Compensation Summary, BLS Employer Costs for Employee Compensation survey, June 2026 reference period, released September 9, 2026. Average for all private industry workers, not specific to receptionists or to small firms. · U.S. Bureau of Labor Statistics, data from 2026