AI receptionist cost comes in three parts: a set-up cost to build and connect it, a usage cost that is usually charged by the minute of talk time, and monthly fees for the phone number, the platform and whoever looks after it. There is no single going rate, because the price moves with your call volume, how long calls run, what the AI connects to and who maintains it, so the useful comparison is what it costs against the work your missed calls lose you now.
A disclosure: CentriWeb builds and looks after voice agents for clients, on platforms of this kind, so weigh our view with that in mind. We do not quote our own prices here, and we have not invented market price ranges. Where we mention a platform’s rates, they come from its own pricing page.
The three ways AI receptionist pricing is built
Set-up
Someone has to write down what the AI should know and say, connect it to your phone number, calendar and CRM, set the rules for handing calls to a person, and test it on real questions. A self-serve tool puts that work on you. A provider that builds it for you charges for it, either as a one-off fee or spread across a contract.
Usage, by the minute
The voice platforms underneath most AI receptionists charge by the minute, and that minute is made of several parts. As at November 2026, Retell’s pricing page breaks a voice agent minute into the voice engine, the language model, the text-to-speech voice and telephony, and gives a worked example of US$0.11 a minute for one combination. Vapi’s pricing page lists a hosting fee of US$0.05 a minute on its pay-as-you-go option, with transcription, model, voice and telephony charged on top at provider cost. Check their pages for current prices, because they change, and note that both are in US dollars.
If you buy an AI receptionist from a provider rather than straight from a platform, you will usually see one per-minute or per-call rate, or a monthly allowance. The platform costs are inside it.
Monthly fees
On top of minutes, expect some of these: a phone number, a platform or plan fee, extra capacity for calls at the same time, and a support or management fee. Retell’s page, for example, lists monthly charges for phone numbers, extra concurrent calls and knowledge bases. Vapi’s page lists monthly plans alongside pay-as-you-go. Our guides to Retell AI pricing and Vapi pricing go through each line.
Self-serve, packaged or built for you: where the money goes
The same AI receptionist can be bought three ways, and each moves cost between the price list and your time.
Self-serve, straight from a platform. You pay the platform’s usage and monthly fees and do the set-up yourself. The bill is lowest, but you write the instructions, connect the calendar, test the calls and fix it when something changes. That suits an owner with the time and some technical confidence.
A packaged AI receptionist product. A provider sells a ready-made receptionist on a monthly plan, often with a call or minute allowance. Set-up is lighter because the product decides most of the structure. It works well when your calls fit what the product was built for, and less well when your booking rules or systems do not.
Built and looked after for you. Someone builds the agent around your calls, connects it to your systems and keeps it current. Set-up and upkeep cost more than a packaged plan, and you hand over most of the time cost. It makes sense when the booking rules are fiddly, when the details must land in a particular CRM, or when nobody in the business has the hours to maintain it.
Whichever way you buy, the usage underneath is still minutes of talk time, so call volume and call length drive the running cost in all three.
What moves the cost up or down
| Cost line | What it covers | What makes it bigger |
|---|---|---|
| Set-up | Writing the knowledge, script and rules; connecting phone, calendar and CRM; testing | Complicated booking rules, several systems to connect, many services or locations |
| Usage | Each minute of talk time across the voice, model and phone parts | More calls, longer calls, a premium voice or model, time spent on hold |
| Platform and number | The phone number, plan fee, capacity for calls at once | Several numbers, high peak volume, extra features |
| Upkeep | Updating information, reviewing calls, fixing what breaks | Frequent changes to prices, staff or services; more connected systems |
Billing rules matter as much as the rate. Retell’s page says, as at November 2026, that calls are tracked to the nearest second, failed calls are not billed, and billing continues during silence or hold. Ask any provider how it treats short calls, transfers, hold time and calls that go to voicemail.
How to work out an AI receptionist cost for your business
You need three numbers from your own phone records: calls a month the AI would take, average length in minutes, and the per-minute or per-call rate in the quote you are comparing. As an example, suppose:
- the AI would take 300 calls a month;
- they average three minutes each, so 900 minutes;
- your quote’s all-in rate is R a minute.
Usage is 900 times R. Add the monthly fees in the quote, then spread the set-up over the months you expect to keep it. If you expect to keep it two years, divide set-up by 24. That gives you a monthly cost you can compare with anything else.
Two checks keep the estimate realistic. Pull a month of real call lengths rather than guessing, because a few long calls change the average. And count after-hours calls separately, since they are often the ones an AI receptionist exists to catch.
Comparing the cost with the work you miss now
The cost on its own tells you little. What matters is whether it saves time, brings in more money or keeps money you already make. Work it out from your own numbers:
- Missed calls a month. Your phone system or carrier can usually show these. Say, as an example, 40.
- How many were new work. Not every missed call is a job. Say half, so 20.
- How many would have booked. Use your own conversion rate from calls you do answer. Say one in four, so five.
- What a booking is worth. Use margin, not the invoice total.
Five bookings times your margin per booking is the monthly work at stake. If that is well above the monthly cost from the last section, the AI receptionist is worth testing. If it is close, a cheaper fix such as a missed-call text back may do the job. These numbers are an example only, so replace every one with yours.
Count time as well. If you or a staff member spends an hour a day on calls the AI could answer, that hour is worth something even if no extra jobs come in.
Costs that do not show on the price list
- Your time at set-up. Someone has to supply prices, policies, service areas and the answers to awkward questions. A self-serve tool needs more of it.
- Keeping it current. A price rise, a new service or a staff change means updating the AI. If nobody does, it gives callers old information.
- Reviewing calls. Reading a sample of transcripts each week is how you catch mistakes early.
- Currency. Platforms that price in US dollars move with the exchange rate.
- Legal set-up. Recording notices and privacy settings take time to get right. Our guide to call recording laws in Australia covers the state rules.
Questions to ask before you accept a quote
- What is included in set-up, and what costs extra later?
- How is a minute or a call counted, including hold time, transfers and very short calls?
- Which monthly fees apply, and do any rise with volume?
- Who updates the AI when something changes, and is that included?
- What does it connect to, and who fixes the connection when a system updates?
- Can we keep our phone number and our call data if we leave?
- Is pricing in Australian or US dollars?
CentriWeb quotes its AI receptionist work after looking at a business’s calls, because volume and booking rules change the price more than anything else.