What an AI Receptionist Really Costs (the Per-Minute Math, in Euros)

Search “AI receptionist cost” and you’ll get ten pricing pages from US vendors, each one concluding that the vendor writing it is the best deal. Fine. But none of them show you what a minute of AI phone call actually costs to produce, and none of them were written for a clinic in Athens or a law office in Lisbon. I build these for a living at Code2b, mostly for European small businesses, so this is the version with the hood open: the raw per-minute parts, what a SaaS plan marks them up to, when a custom build makes more sense, and the EU costs that US guides never mention.

Short answer: The raw cost of an AI receptionist call in 2026 is roughly $0.07 to $0.30 per minute. Off-the-shelf plans turn that into $25 to $300 a month depending on volume. A custom build that talks to your own booking system and CRM usually lands at a one-off setup of €2,000 or more plus €200 to €500 a month. Which one is cheaper for you depends almost entirely on two numbers: how many minutes you take, and how many systems the receptionist needs to touch.

The four meters running on every call

Every AI receptionist, whatever the logo on the dashboard, is the same four services chained together. Each one bills by the minute (or by the token, which works out to about the same thing).

1. Telephony. The phone number and the actual call leg. Around $0.01 to $0.02 per minute, plus a small monthly fee per number. European numbers cost a bit more than US ones, and some countries need proof of a local address before you get one.
2. Speech-to-text. Turning the caller’s voice into words. This is the cheapest part now, well under a cent per minute on the good engines.
3. The language model. The brain that decides what to say. This is the part with the widest range. A small fast model can cost a fraction of a cent per minute. A top model on a chatty call can pass $0.15.
4. Text-to-speech. The voice. Cheap synthetic voices start around $0.015 per minute. The premium, very human ones run $0.04 to $0.10.

On top of that, if you use a voice platform to glue it together (Vapi, Retell and similar), there’s an orchestration fee of about $0.05 per minute. Retell’s public price sheet puts the whole stack at $0.07 to $0.31 per minute depending on the model and voice you pick. That matches what I see on real invoices.

Here’s what that looks like in practice:

The thing to notice: even the premium stack costs less per minute than a phone call to most mobile numbers used to cost a decade ago. The minutes are not where the money goes.

So why do plans cost $199 a month?

Because the raw minutes are the smallest part of the bill. A SaaS receptionist at $199 a month for “unlimited” calls is selling you three things on top of the minutes:

• Someone else’s setup work. Prompts, call flows, voicemail fallback, testing.
• A dashboard and support. Call logs, transcripts, a person to email when it breaks.
• Margin on the average customer. Most small businesses take far fewer calls than they think, so a flat price covers the heavy users with the light ones.

That’s a fair deal for a lot of businesses. If you’re a plumber in Ohio who needs calls answered, a message taken and an appointment dropped in Google Calendar, buy the SaaS plan. Don’t pay anyone (including me) to build it.

Where off-the-shelf stops working

The standard advice (and I give it too) is to start with a SaaS tool and only build custom once you outgrow it. But after enough calls with clinics, salons and professional firms in Greece, I noticed the same four walls come up again and again, and they come up in week one, not year two.

The language. Most SaaS receptionists are built English-first. They’ll list Greek, Portuguese or Czech as “supported”, and technically they are. Then a caller says a street name, a surname or a medical term, and the transcription falls apart. Local-language quality depends on which speech engine and which voice the vendor picked, and you usually can’t change either.

The booking system. The SaaS tools integrate beautifully with Google Calendar, Calendly and the big US practice-management systems. They don’t integrate with the dental software your clinic has used for twelve years, or the salon booking system that’s popular in your country and nowhere else. Without that link, the AI takes a message instead of booking, and you’ve paid for a very polite voicemail.

Knowing who’s calling. A good human receptionist says “Hi Maria, calling about Thursday?” before Maria finishes her name. That needs a lookup against your customer database while the phone is ringing. Generic tools can’t do it unless your data lives somewhere they already support.

Data location and compliance. More on this below, because it’s the part that actually costs money.

If none of those four apply to you, go SaaS. If two or more do, the “cheap” plan is going to cost you in missed bookings and workarounds, and the math changes.

The European part nobody prices in

Every US pricing guide I read while writing this skipped Europe entirely. Here’s what’s different.

You have to tell callers they’re talking to an AI. Since 2 August 2026, Article 50 of the EU AI Act requires AI systems that talk to people to make that clear, unless it’s obvious from context. For an AI phone line, that means the greeting says so. This costs nothing to build, but some tools let you give the AI a human name and persona with no disclosure at all, and in the EU that’s now a compliance problem, not a clever feature. (Callers mostly care that the call gets handled, not who handles it.)

GDPR still applies to every recording and transcript. A call transcript contains a name, a phone number and often a reason for the visit, which at a clinic is health data. You need a data processing agreement with every provider in the chain, you need to know where the audio is stored, and you need a retention rule so transcripts don’t pile up forever. Many US tools will sign a DPA if you ask. Fewer can tell you which country each of the four components runs in.

Phone numbers. Getting a local number in some EU countries needs a registered address and ID documents, and number porting from your current carrier can take a couple of weeks. Budget for the delay, not just the fee.

None of these make an AI receptionist a bad idea in Europe. They just mean the true cost includes an hour or two of someone checking the paperwork, and that “cheapest plan wins” is the wrong way to choose.

The worked example: a two-dentist clinic

Numbers make this concrete, so here’s a hypothetical clinic I’d call typical of the businesses we talk to. Two dentists, one front desk, around 400 inbound calls a month averaging two and a half minutes each. That’s about 1,000 minutes a month. Roughly a third of calls come in while the front desk is busy with a patient or after hours.

Option 1: hire a second receptionist, part-time. The Greek minimum wage is €920 gross a month since April 2026. Paid 14 times a year with employer contributions of around 22% on top, a full-time person costs the business roughly €1,300 a month at minimum wage, and experienced front-desk staff cost more. A part-timer covering evenings is still €650 or more, and they cover maybe 20 of the 168 hours in a week.

Option 2: a US SaaS receptionist. Around $199 a month flat, so under €200. It answers every call. It takes messages well in English and adequately in Greek. It can’t see the clinic’s appointment book, so it can’t actually book, and someone calls every lead back the next morning. Real cost: the subscription plus 30 to 45 minutes of the front desk’s day, plus the patients who booked elsewhere before that callback happened.

Option 3: a custom-built receptionist. Raw usage at a mid-range stack is about 1,000 minutes × $0.15, so roughly $150 a month in API and telephony costs. Add hosting, monitoring and maintenance, and a build like this usually sits at €200 to €350 a month, plus a setup fee from about €2,000 that covers connecting it to the clinic’s booking software. It answers in natural Greek, checks the real calendar, books the slot, sends an SMS confirmation and flags anything clinical for a human.

Over the first year, option 2 is the cheapest line on the spreadsheet. Option 3 costs more in month one and less by month six, because every booked appointment it catches at 9pm is revenue option 2 hands back to the front desk as a to-do. If the clinic’s average visit is worth €60 to €80, catching five or six extra bookings a month pays the difference.

That’s the whole decision, really. You’re not comparing prices. You’re comparing how many calls end in a booking.

What actually drives the price of a custom build

If you go the custom route, these are the things that move the quote, roughly in order of impact:

• Number of integrations. The first one (usually the calendar or booking system) is the big one. Each additional system, like a CRM, billing tool or customer database, adds setup work and a small monthly cost to maintain. At Code2b we include one integration in the setup and charge €90 a month for each extra.
• How far the AI is allowed to act. Answering questions is cheap. Booking is more. Rescheduling, cancelling and taking deposits each add logic and testing, because a mistake there costs a real customer.
• Call volume. Matters less than people think, because raw minutes are cheap. Doubling calls from 1,000 to 2,000 minutes adds maybe €130 a month.
• Languages. Two languages is not twice the work, but each one needs its own testing with real accents and real local names.
• Escalation rules. When should it transfer to a human, who gets the transfer, and what happens if nobody picks up? Getting this right is most of the difference between a demo and something you’d trust on a Saturday. I wrote more about that gap in the production gap.

For the broader picture of how we price automation work in general, see what it costs to automate a business process.

Where to go from here

Before you compare a single plan, pull last month’s call log from your phone provider and count two things: total inbound minutes, and how many calls went unanswered. That’s your real budget input. If unanswered calls are under 5%, you probably don’t need an AI receptionist at all. If they’re above 20%, the cheapest option is whichever one books the most of them.

If you want a second opinion on your numbers, we build these at Code2b as an AI voice receptionist service, priced on our pricing page. Or book a 15-minute call and bring the call log. I’ll tell you straight if SaaS is the better fit.

Questions people ask before signing anything