MarketStarter / System 01 · Voice AI / Cost
What an AI receptionist really costs.
Every vendor in this category quotes a headline rate and bills the rest on top. So here is the whole stack, priced from the line on our own desk — the telephony, the voice, the model — and then measured against what the market charges for the same answered call.
The short answer
An AI receptionist costs roughly €0.072 per minute of conversation to run, on top of a phone line that costs about €19 a month whether it rings or not. At 300 minutes a month that is around €41 of direct cost.
Bought as a managed service, expect €290–€690 per month for 300–1,500 minutes, plus a one-off €1,490–€2,490 to install and integrate it. Those are MarketStarter's own published numbers, and the rest of this piece is where they come from.
For comparison, human answering services publish $3.45–$5.00 per minute (Ruby) and $7.00–$11.50 per call (Smith.ai). At comparable volumes an AI line is about three to five times cheaper per minute — with a real caveat about what you give up, in rent vs own.
Why nobody publishes this number
Search for what a voice agent costs per minute and you get three figures: Vapi at $0.05, Retell at $0.07, Bland at $0.09. None of them is what you pay. Each is an orchestration fee, and speech-to-text, the language model, the voice synthesis and the telephony are billed separately on top — the vendors say so in their own documentation. Published all-in costs across the same platforms land between $0.07 and $0.50 per minute, a seven-fold spread hiding behind a two-cent headline.
Above them sits managed AI: Synthflow's enterprise tier starts at $30,000 a year, scoped on volume, concurrency, telephony and integrations. And above that sits the market these products are actually replacing — people answering phones — where the pricing has been public and stable for years.
What follows is the version nobody has an incentive to publish: every line item of a working Greek business line, from a real invoice, with the estimates labelled as estimates.
What the parts actually cost
Four layers, and only two of them cost real money.
1 · The line — about €19 a month, before anyone calls
The most recent invoice for the SIP line behind our own agent reads €19.25 for the month, of which €0.27 was actual call traffic. Everything else is subscription. That is the shape of telephony: a fixed floor per line, plus per-minute charges that only matter at volume.
It is also the number nobody quotes, because carriers in this market don't publish per-minute rates for Greek numbers — you request a price list by email. That opacity is worth knowing about before you sign anything: it is the one line item in this stack you cannot verify from a public page.
2 · The PBX — €0, and that is the interesting part
Phone-system vendors sell AI add-ons that require a paid annual subscription. We don't buy them. In our architecture the PBX does one job — forward the call to the agent when the team is offline or every line is busy — and the free 3CX edition does that at no cost. The agent itself lives outside the phone system entirely.
You are not buying your phone vendor's AI. You are putting your own agent on top of the phone system you already have. That is the same ownership argument that runs through the rest of this piece, one layer down in the stack.
3 · The voice — $0.080 a minute, on every single tier
This is the expensive layer, and it has a property that changes how the whole thing should be priced. On ElevenLabs, the conversational voice tiers work out to exactly $0.080 per minute at every level. The subscription is not a discount — it is prepaid minutes at the same unit rate.
| Plan | $ / month | Minutes | Effective $ / min | Concurrent calls |
|---|---|---|---|---|
| Starter | 6 | 75 | 0.080 | 6 |
| Creator | 22 | 275 | 0.080 | 10 |
| Pro | 99 | 1,238 | 0.080 | 20 |
| Scale | 299 | 3,738 | 0.080 | 30 |
| Business | 990 | 12,375 | 0.080 | 40 |
Divide any row and you get $0.080 to the cent. What the higher tiers buy is concurrency — how many calls can be in progress at the same moment — and nothing else. Exceeding your concurrency does not queue the call; it bills at the burst rate of $0.160 per minute, double.
A dental clinic with 300 minutes bunched into Monday morning costs materially more than 300 minutes spread across a month, because the bunched version burns burst minutes. Any price built only on minutes leaves that risk unpriced — which means it eventually arrives as a surprise on someone's invoice.
4 · The model — about $0.004 a minute, or 5% of the bill
Intuition says the AI is the expensive part. It isn't, and it isn't close.
A conversation minute runs roughly 2.7 turns. With a system prompt and a growing transcript held in cache, that is on the order of 3,500 input tokens and 160 output tokens per minute. At Claude Haiku 4.5's published rates that comes to about $0.0043 a minute; at Sonnet 5's, about $0.013.
The voice costs about nineteen times more than the brain. Two things follow. First, upgrading to the stronger model adds roughly a cent a minute — quality is effectively free, so there is no good reason to run a customer-facing agent on the cheapest model available. Second, any real cost saving has to come out of minutes, not tokens. Prompt golfing is a rounding error; a shorter, better-structured call is not.
The monthly cost curve
Put the layers together and the shape is simple: a fixed floor of about €19, plus about €0.072 for every minute of conversation.
| Minutes / month | Voice | Model | Line | Direct cost |
|---|---|---|---|---|
| 100 | €6.9 | €0.4 | €19 | ≈ €26 |
| 300 | €20.6 | €1.1 | €19 | ≈ €41 |
| 600 | €41.3 | €2.2 | €19 | ≈ €63 |
| 1,000 | €68.8 | €3.7 | €19 | ≈ €92 |
| 2,000 | €137.6 | €7.4 | €19 | ≈ €164 |
Excludes per-minute telephony, which our carrier does not publish; on any plausible rate it moves the totals by single-digit euros at these volumes. Note what is missing: an economy of scale. Only the €19 floor gets amortised. The variable cost is flat forever, which is exactly why the vendors' own subscriptions are flat too.
What the market charges for the same answered call
The real competition for an AI receptionist is not another AI receptionist. It is a company that employs people to answer your phone, and those companies publish their prices.
| Plan | Minutes | $ / month | $ / minute |
|---|---|---|---|
| Starter | 50 | 250 | 5.00 |
| Standard | 100 | 395 | 3.95 |
| Professional | 200 | 720 | 3.60 |
| Enterprise | 500 | 1,725 | 3.45 |
Smith.ai prices the same job per call rather than per minute: $300 for 30 calls ($10.00 each), $810 for 90 ($9.00), $2,100 for 300 ($7.00), with overage at $8.50–$11.50 and no setup fee.
Two things are worth reading out of the Ruby table. The first is the absolute level: a small business is already paying $250–$1,725 a month for someone to pick up the phone. Nobody has to be convinced the problem is worth money — the category is priced, public and old.
The second is the slope. Ruby's per-minute price falls only 31% between the smallest and largest plan, because labour does not get cheaper with volume. Our variable cost is flat too — but it sits at roughly one fortieth of their cheapest published minute. That is the whole structural difference, and it means volume tiers can be generous in a way a staffed service simply cannot match.
| Monthly volume | Human answering service | AI receptionist | Difference |
|---|---|---|---|
| ~200–300 min | Ruby Professional — 200 min, ≈ €619 (€3.10/min) | Starter — 300 min, €290 (€0.97/min) | ≈ 3.2× |
| ~500–750 min | Ruby Enterprise — 500 min, ≈ €1,484 (€2.97/min) | Growth — 750 min, €490 (€0.65/min) | ≈ 4.6× |
| 1,500 min | not published at this volume | Scale — 1,500 min, €690 (€0.46/min) | — |
Per-minute comparison at the closest published volumes, which is the fair way to read it — comparing our largest tier to their smallest would show 9× and mean nothing. Setup is excluded from both columns: Smith.ai charges none, Ruby publishes none, and we charge €1,490–€2,490. That gap is addressed below.
What we charge, and why
The pricing follows the cost structure rather than the market's habit. Because the cost has two drivers — minutes and concurrency — the price has two drivers as well.
System 01 · Voice AI Calling Agent
Monthly, VoIP line and voice usage included
- Starter€290300 minutes · 2 concurrent calls
- Growth€490750 minutes · 4 concurrent calls
- Scale€6901,500 minutes · 6 concurrent calls
Installation: €1,490 for one line, one language, a calendar and one CRM — up to €2,490 for bilingual with outbound calling and two integrations.
Minutes beyond the tier: €0.35 each. An extra concurrent line: €45 per month.
Prices exclude VAT. What shapes your number is only ever the five scope drivers on the system page — lines, languages, direction, integrations, volume.
Three things about those numbers are deliberate.
- Concurrency is priced, not assumed. Every tier names how many calls can run at once, because that is what the underlying platform actually sells. The alternative — quoting minutes only — hands the burst risk to whoever gets surprised first.
- Overage is €0.35, not €0.072. Going over a tier is not just extra platform cost; it is the signal that the setup is sized wrong and needs revisiting. The gap is deliberate and it is cheaper to move up a tier, which is the point.
- Installation is a real cost, not a formality. Twelve to twenty hours go into the line, the integrations, the prompt work and the tuning. It is charged once because it happens once — and because it is the part you keep.
Rent or own — including the part that argues against us
Ruby and Smith.ai sell a subscription to their people. That is a coherent product and in some situations it is the better one. What it is not is the same category as this: when you stop paying, you stop having a receptionist, and the number, the scripts and the call history stop being available to you.
Here, the agent is installed on your own line and your own 3CX, with your prompts, your integrations and your call data. You keep all of it. The monthly buys maintenance, tuning and the platform usage — not permission for the system to keep working.
We cost more up front, and "low monthly, cancel anytime" is a genuine advantage. Smith.ai charges no setup fee at all. If you want to try answering-service coverage for one quarter, renting is cheaper and faster, full stop. Ownership only pays back over a horizon long enough to absorb €1,490–€2,490 — and it should not be sold as "pay once and it's free forever", because there is still a monthly.
The other honest difference: they staff humans. For calls that need judgement, empathy or improvisation on the first attempt, a person is better than an agent, and no per-minute table captures that.
So the decision is not really about price. It is about whether the phone is a cost to cover or a system you want to own — and about volume, because the arithmetic only works one way past a few hundred minutes a month.
When an AI receptionist costs more than it saves
Three cases, said plainly, because a cost article that only argues one direction is an advert.
- Very low call volume. Under roughly 40–50 calls a month, the fixed monthly and the installation are hard to justify against what a missed call is actually worth to you. Voicemail and a disciplined callback habit are cheaper.
- Every call needs a human decision. If the first minute of every conversation requires judgement that can't be written down — bespoke quoting, clinical triage, crisis work — an agent is a filter at best, and you should be pricing a person.
- Nothing to forward from. If calls arrive on a personal mobile with no VoIP or PBX in front of them, the phone system has to be sorted out first, and that is a separate project with its own cost.
The case where it does pay is narrower than the marketing suggests and more common than owners expect: a business where the phone drives revenue and a measurable share of calls goes unanswered — clinics mid-treatment, agencies with everyone on viewings, workshops after closing, anyone whose callers simply dial the next name on the list.
What it costs when the agent calls out
The per-minute engine is identical. The cost driver is not.
On inbound, minutes are set by how many people call you. On outbound, minutes are set by how many people pick up — not by how many meetings get booked. A prospect who answers, listens for ninety seconds and declines costs exactly as much as one who books. That single fact is why pure pay-per-meeting pricing is a trap for whoever offers it: it puts the entire connect-rate risk on one side of the table, and that risk gets priced back in eventually.
The published market agrees. B2B appointment-setting retainers run $3,000–$10,000 a month; the cheapest published cold-call-only programme is $3,500. Per meeting, the bands are $150–$300 for basic qualification, $300–$500 for meetings matched to the client's own criteria, and $400–$750 where no-show replacement is included. Hybrid pricing — a retainer plus a per-meeting fee — is a normal structure, not an unusual one.
System 06 · Appointment Setting
€2,490 per month, plus €150 per attended qualified meeting
Meetings are charged on attendance — a no-show that reschedules and then happens counts once; a meeting the client cancels counts. There is a monthly ceiling at the target +20%, and anything beyond it needs written approval. Month one is setup with no target, and if after two months the segment isn't producing, either side exits on 30 days' notice with no penalty.
That puts us roughly 17% below the cheapest published competitor in the world, against a market where meetings qualified to the client's criteria list at $300–$500. Details on the system page.
What the rest of the systems cost
The same principle applies across the catalogue: where the cost driver is usage, the price scales with usage; where it is our hours, the scope is capped in hours so the number stays honest. All figures exclude VAT.
System 02 · AI Automation
€590–€990 per workflow (scoped to a maximum of four hours of build), or €1,490–€3,900 for a project. Care and monitoring afterwards: €190–€490 a month.
The runtime is almost free — we run n8n self-hosted, where the vendor's own comparable cloud tier is €50 a month, and a single small server carries every client at once. So what you are buying is build hours, which is exactly why the per-workflow scope is capped at four of them rather than left open. System 02 →
System 03 · AI SEO / AEO
€890 for a one-off audit, then €890 / €1,490 / €2,290 a month with a defined hour ceiling per tier.
Tooling here costs cents: scoring 2,000 keywords for difficulty and intent and reading thirteen full result pages — the research pass behind our own content plan — cost $0.56 in total, and Search Console and Bing Webmaster Tools are free. Effectively all of the cost is research, writing and schema work. For context, the Greek market prices a GEO audit at €600–€1,500 and combined SEO+GEO retainers at €700–€2,500 a month. System 03 →
System 04 · Cold Outreach Engine
€890 setup — domain warm-up, ICP definition, sequences — then €590–€990 a month per roughly 1,000 contacts.
The binding constraint is data credits, not seats: a research seat carries a fixed monthly credit allowance, and a thousand contacts consumes roughly a quarter of it — so four campaigns running at once exhaust the plan, regardless of how many people are logged in. That is why the price is quoted per contact volume rather than per user. System 04 →
System 05 · CRM Quick-Start
Fixed: €1,900 Starter · €2,900 Pro · €4,900 Advanced. Optional admin retainer from €290 a month.
Three fixed prices instead of an hourly rate, because the scoping conversation was costing more than the difference. Each tier is a named list of deliverables drawn from a fixed menu of quick wins — pipelines, inbox sync, properties, automations, dashboards — so what moves the price is which tier you pick and nothing else. System 05 →
System 09 · AI Performance Ads
€990–€1,490 to build — tracking, conversion wiring and the campaigns — then €590–€890 a month to run, or about 12% of ad spend once the budget is large enough to make that the bigger figure. €590 is the floor either way. The advertising budget itself is never ours: it is paid on your card, in your accounts.
There is no per-minute meter here, so unlike the phone line the entire cost is hours — roughly eight to twelve to build an account properly and four to six a month to keep it honest. That is also why the floor exists. At €390 a month the work is being done for nothing, and an account nobody can afford to look at is worse than no account, because it keeps spending either way. The corollary is the number we give away on the system page: below roughly €1,000 a month in media the fee is a bigger line than the advertising, and at that level you should take the build and run it yourself. For context, Greek SMEs typically run €300–€1,500 a month in ad budget. System 09 →
What the management half actually consists of is set out on a real account in six months of a school’s Google Ads, month by month — including the month-by-month cost-per-enquiry table that is the cheapest analysis in this business and the one almost nobody runs.
System 10 · E-shop Launch System
€3,900–€5,900 one-off for the entry build — store, design pass, product data, email flows and the tracking stack — with anything beyond that list quoted as its own piece. Optional growth care afterwards: €290–€590 a month. The platform is not ours to bill: the Shopify subscription, any paid apps and the domain are paid by you, directly, and typically land between €30 and €90 a month.
This is the one price in the catalogue that went up when we checked it. We had been quoting nearer €2,900, and against the hours a real build actually takes that left a margin of about 3% — the only service on the menu that was, in effect, losing money politely. Publishing that is less embarrassing than the alternative, which is a floor nobody can explain. It is also why the cheap end of this market works the way it does: at €500–€700 a store is a theme installed on a domain, and it has to be, because there are no hours in that number for flows, measurement, a product feed or imagery. Those are not features somebody withheld from you — they are the entire build. For context, a professional custom build in Greece runs €2,500–€8,000. System 10 →
System 11 · E-shop AI Support Concierge
€1,690–€2,900 to build, then €190–€490 a month by conversation volume. Adding the phone to the same brain is System 01's Starter tier at €290 a month, carrying System 01's own meter of minutes — voice is expensive in a way text is not, and it is never folded into the conversation count.
This one now has an article of its own, so the detail lives there rather than being repeated here: the three named volumes, the overage, what counts as a conversation, and what a conversation actually costs to run. It also carries the part that decides whether to buy it at all — the message volume below which automation does not pay for itself, and why a containment rate is the wrong number to compare suppliers on. What customer service automation can and can't answer → System 11 →
Scoped builds
CRM Implementation from €4,900 one-off plus €490–€1,490 a month; Inbound / Content Engine €1,490–€3,900 a month. These two are quoted from a call rather than a tier, because the work genuinely changes with the shape of the business — seats, pipelines, integrations and data complexity for the implementation; how much gets published, in how many languages, and whether the measurement already exists for the engine. This is where the complexity is allowed to live, on purpose. The implementation is delivered in phases with a fixed price per phase, and the engine is a named monthly list rather than an open retainer, so in both cases the open-ended part is the scope, never the meter. System 07 → System 08 →
Sources, and how confident we are in each number
Every figure above carries an evidence grade, because "our costs" and "a number we read on a blog" are not the same claim.
- Our own invoices — the €19.25 line, the €0.27 of traffic, the platform subscriptions we actually pay. Strongest evidence here; also the only numbers in this article that no competitor can publish, because they are ours.
- Vendor price pages — the ElevenLabs tiers, the 3CX free edition, the self-hosted automation runtime. Public and checkable, as of July 2026.
- Competitor price pages — Ruby and Smith.ai's published plans, the developer platforms' headline rates, Synthflow's stated enterprise floor, and the appointment-setting benchmarks. Public as of July 2026, and all in USD.
- Calculated — the per-minute model cost and the monthly cost curve. Assumptions are stated inline: 2.7 turns a minute, ~3,500 input and ~160 output tokens, published model rates, and $1 ≈ €0.86. Change the exchange rate or the model and these move; nothing else here does.
A word on shelf life. Prices in an article age badly and get quoted back for years. This page carries a visible review date at the top and is checked when the underlying costs move — the exchange rate, the platform tiers, or our own line. If you are reading it long after , treat the competitor figures as historical and ask us for the current ones.
MarketStarter is a HubSpot Solutions Partner and a Salesforce Partner, based in Greece, answering everywhere.
Straight answers
The cost questions, answered.
How much does an AI receptionist cost per minute?
The raw platform cost of a conversation minute is about $0.084, roughly €0.072: $0.080 for the voice layer and about $0.004 for the language model, with telephony billed on top. Sold as a managed service, a minute retails for €0.46 to €0.97 depending on the monthly volume it sits inside. The gap between those two numbers is installation, integration, prompt engineering and maintenance — not margin on the minute itself.
What does an AI receptionist cost per month?
At MarketStarter, €290 per month for 300 minutes and 2 concurrent calls, €490 for 750 minutes and 4, or €690 for 1,500 minutes and 6 — VoIP line and voice usage included. Minutes beyond the tier are €0.35 each and an extra concurrent line is €45 per month. Installing the agent is a separate one-off of €1,490 to €2,490 depending on languages, direction and integrations. Prices exclude VAT.
Is an AI receptionist cheaper than a virtual receptionist service?
On published rates, yes — by roughly three to five times at comparable volumes. Ruby publishes $250 for 50 minutes up to $1,725 for 500 minutes, which is $3.45 to $5.00 per minute; Smith.ai publishes $7 to $11.50 per call. A 300-minute AI tier at €290 works out at €0.97 per minute. The honest caveat is that those services staff real people, which is a genuine difference in what the caller gets, not just a difference in price.
Why do AI voice platforms advertise $0.05 per minute?
Because the headline rate covers orchestration only. Speech-to-text, the language model, the voice synthesis and the telephony are each billed on top, and the vendors say so in their own documentation. Published all-in costs across the developer platforms land between $0.07 and $0.50 per minute. Any quote that does not name the telephony carrier and the model is incomplete.
Does a bigger plan make each minute cheaper?
No — and this is the most misunderstood part of the economics. On ElevenLabs, every tier from Starter to Business works out at exactly $0.080 per conversation minute; the subscription is prepaid minutes at that rate. What the higher tiers buy is concurrency — how many calls can be in progress at once, from 6 up to 40. There is no volume discount on the minute, so a price built only on minutes leaves the burst risk unpriced.
What is the cheapest an AI receptionist line can cost to run?
About €19 a month, before anyone calls. That is a real invoice from our own business line: €19.25 for the month, of which €0.27 was actual traffic. The PBX layer can be €0, because a free 3CX edition is enough to forward calls to the agent. Everything above that floor is minutes at roughly €0.072 each. That is the running cost, not the price of a working setup — installation, integrations and prompt work are what a business is actually buying.
What does it cost when the AI calls out instead of answering?
The per-minute engine is identical, but the cost driver changes: outbound minutes are set by how many people pick up, not by how many meetings get booked. Published B2B appointment-setting retainers run $3,000 to $10,000 a month, with $150 to $300 per basic meeting and $300 to $500 for meetings qualified against the client’s own criteria. MarketStarter charges €2,490 per month plus €150 per attended qualified meeting.
Do I need to buy a new phone system?
No. The agent answers over VoIP on a new line, or on your existing number by forwarding — no porting. In our architecture the PBX costs nothing, because the free 3CX edition does the forwarding and the agent itself lives outside it. You are not buying your PBX vendor’s AI add-on; you are putting your own agent on top of the phone system you already have.