What an AI receptionist really costs a five-person service business

The missed-call problem is a cost question

For a five-person service business — plumbing, HVAC, salon, clinic — the trigger is familiar: everyone is on a job site, the phone rings, voicemail fills, and the lead goes elsewhere. The question this post answers is not whether an AI receptionist works.

A crew that size lives on first-call capture. When the phone rolls to voicemail and voicemail sits unopened until evening, the cost shows up as lost bookings, not as a line item. Vendors pitch the fix as obvious; the harder question is what the fix costs each month and what you surrender when you sign.

Framed honestly, the decision is cost math plus lock-in risk. Based on published pricing pages, a 900-minute month (300 calls at three minutes each) comes to roughly $103–$145 all-in on a Vapi path and about $99–$315 on a Retell path, including the n8n glue layer. Synthflow publishes no figure at all. The layer you can cancel matters at least as much as the total.

Basis of this post. Everything below is based on published pricing pages and vendor documentation from Vapi, Retell AI, Twilio, Synthflow, and n8n, reviewed on 2026-10-30. No calls were placed, no platforms were operated, and no agents were built for this article. Every price and retention figure links to the page it came from, and where no published figure exists, the post says so instead of estimating one.

What an AI receptionist does

An AI receptionist is two pieces, not one. A voice agent answers the call, listens, captures intent, and decides whether to book, route, or take a message. A workflow layer then acts on that outcome: it writes to a calendar, updates a CRM, or sends an SMS. The split matters because the layers carry separate costs and separate lock-in.

The voice agent — sold by platforms such as Vapi, Retell AI, or Synthflow — handles the conversation: it greets the caller, asks what is needed, and either books, routes, or takes a message. The workflow layer, where a tool like n8n fits, does the acting: writing the booking to a calendar, pushing a note into a CRM, firing off a confirmation text.

n8n counts one execution as one full workflow run regardless of step count, which is why a 300-call month is a small number for that layer. What this post deliberately does not do is claim accuracy, answer rates, or containment figures for any of these agents — no such numbers appear on the pricing pages, and none are invented here.

The three-layer cost stack

Every bill for an AI receptionist is really three bills stacked together. Telephony covers the phone number and call transport. The voice-AI platform covers the agent that talks. The glue layer connects calls to calendars, CRMs, and messaging. Vendors quote only their own layer, so a headline price is never the total.

Layer 1: telephony. Twilio’s US voice pricing lists a local number at $1.15/month and a toll-free number at $2.15/month, with inbound local calls at $0.0085/minute, outbound US and Canada calls at $0.0140/minute, recording at $0.0025/minute, and transcription at $0.05/minute. Twilio also offers a $15 free trial credit. This layer exists whether or not any AI is involved.

Layer 2: the voice-AI platform. Vapi’s pricing page charges $0.05/minute for hosting on its pay-as-you-go package, with models passed through at cost, and it is explicit that transport fees — Twilio inbound at $0.008/minute, Telnyx at $0.0055/minute — are charged by the telephony provider, not by Vapi. Retell AI’s pricing page quotes $0.07–$0.31/minute for AI voice agents, lists US telephony on Twilio numbers at $0.015/minute, and charges $2/month for a Retell-provisioned phone number (custom SIP numbers carry no charge). Synthflow’s pricing page publishes no per-minute rate at all: every deployment starts on an annual enterprise agreement.

Layer 3: the glue. n8n’s pricing page lists a Starter plan at $20/month billed annually, which is an annual commitment to budget as a fixed cost, with 2.5K workflow executions.

How the layers bill each other deserves a close read before comparing totals. Retell’s own FAQ states that it charges only for connected calls, that billing continues through silence and hold, and that when a call is transferred the AI billing stops but telephony billing continues. A transfer that saves the caller time can still run up the telephony line.

A worked estimate: 300 calls at 3 minutes each

Assume 300 calls per month averaging three minutes each, or 900 minutes. That is an illustrative assumption, not a measurement; substitute your own volume before trusting any figure below. Each number here is either a published per-minute rate or a linear scaling of a vendor’s own published estimate. The $74–$116 Vapi band, for example, is Vapi’s 1,000-minute estimate of $82–$129 multiplied by 0.9.

Layer	Vendor	Published basis	Illustrative monthly cost at 900 minutes
Telephony floor	Twilio	$1.15/mo local number + $0.0085/min inbound	$8.80 ($1.15 + $7.65) — floor, not the AI cost
Voice-AI platform	Vapi	$0.05/min hosting; models, voice, transcriber pass-through at cost	$45 hosting; roughly $74–$116 all-in, linear-scaled from Vapi’s 1,000-minute estimate of $82–$129
Voice-AI platform	Retell	$0.07–$0.31/min pay-as-you-go; page example $0.150/min	$63–$279 platform ($135 at the $0.15/min example), plus $13.50 US telephony at $0.015/min and $2/mo for a Retell-provisioned number
Voice-AI platform	Synthflow	Annual enterprise agreement only; no self-serve price	No published figure exists to list
Glue layer	n8n	Starter $20/mo billed annually; 2.5K executions	$20; 300 executions used of 2,500
All-in, 900 minutes	—	Platform + telephony + glue, from the rows above	Vapi path ≈ $103–$145 ($74–$116 platform + $8.80 Twilio + $20 n8n); Retell path ≈ $99–$315 ($63–$279 platform + $13.50 telephony + $2 number + $20 n8n); Synthflow not computable from published pages

Three details decide whether the real bill lands low or high. First, transport fees stack: Vapi lists Twilio inbound transport at $0.008/minute and Telnyx at $0.0055/minute, charged by the provider, while Retell lists its US Twilio telephony rate at $0.015/minute — so the $8.80 telephony floor is not always the whole telephony line. The table uses Twilio’s $0.0085/minute; Vapi’s $0.008 would shave under $0.50 off 900 minutes. Second, model choice swings the platform bill heavily: Retell lists GPT 6 Luna and Claude 5.5 Haiku at $0.0032/minute, while its page-marked recommended GPT 5.6 Terra sits at $0.064/minute — a twentyfold difference in the model line alone. Third, concurrency: Vapi charges $10/month per concurrent line beyond the package default, and Retell includes 20 free concurrent calls before charging $8/month per additional one.

Two calibration notes. Vapi’s own estimate table puts 1,000 minutes at $82–$129/month including hosting, transcriber, model, and voice lines; the $74–$116 band above is that range linearly scaled to 900 minutes, and it remains an estimate rather than a quote. Retell’s estimator example uses $0.150/minute, which the page illustrates as $15/month at 100 minutes. On the glue layer, 300 calls at one execution per call is 300 executions against n8n Starter’s 2.5K cap — enormous headroom, and the reason the glue layer is rarely the bill that surprises anyone.

Lock-in and data-retention comparison

All three voice-AI platforms are vendor-hosted services, so leaving means exporting data and rebuilding prompts elsewhere. Retention windows differ by package, and the glue layer is the one component you can own.

Retention is where the packages diverge most. Vapi lists 14-day raw data retention on its free package, 30 days on Core at $29/month, and 180 days on Pro, which itself carries a $999/month minimum billed as 10% of the hosting fee. Retell lists custom retention — and HIPAA/BAA terms — only under Enterprise, with nothing published below that tier. Synthflow goes further still: no free plan, no trial, no per-minute rate, and every new deployment begins on an annual enterprise contract, which makes it the deepest commitment of the three by structure.

The escape hatch is the glue layer. n8n’s Community Edition is self-hostable and available on GitHub, so the workflow connecting calls to your calendar and CRM is the one part of the stack that can run on your own infrastructure. n8n’s paid self-hosted tier (Business, 667€/month billed annually) is aimed at companies under 100 employees that want SSO, queue mode, or Git version control. A pragmatic architecture for a lock-in-averse business is a usage-based voice platform with no annual contract, plus a self-hosted n8n instance holding the business logic.

On service commitments, Vapi lists a two-business-day response SLA on Core and a 99% uptime SLA on Pro; the other pricing pages reviewed list no comparable self-serve terms.

When a human receptionist still wins

No published pricing page answers whether an AI agent handles your callers better than a person, and this post will not invent one. What the numbers do show is that fixed fees weigh most at low volume: at twenty calls a month, the $20 n8n plan and number fees make up most of the bill. No human-receptionist cost is sourced here, so this post claims no break-even point.

A shop taking twenty calls a month pays little on pay-as-you-go platforms, and its fixed floor is the $20 n8n plan plus a number fee, so at that volume the stack is a question of service, not of cost. Consider it when the calls themselves demand judgment: a distressed client, a disputed invoice, a medical question needing empathy, a dispatch decision that depends on knowing which technician is having a bad day. None of the pricing pages reviewed claim their agents handle those conversations well, and no sourced data here suggests they do.

The honest framing is situational, not comparative. Where every call is a relationship, the math starts from a different place — and no ROI figure in this post should be read as saying otherwise, because none exists in the sources.

Verdict by business situation

There is no single winner here, only fits. The cheapest credible entry points are usage-based platforms with free credits, while compliance requirements push you toward contract pricing. The right choice depends on call volume, model taste, and how much owning the glue layer matters to you.

Testing or low volume: start where entry is free. Vapi’s $0-to-start package includes $5 in free credits, one phone number, and four concurrent calls. Retell offers $10 in free credits and 20 free concurrent calls. Twilio adds a $15 trial credit for the telephony layer. Synthflow offers no trial at all.
Predictable volume, cost-sensitive: Retell’s $0.07–$0.31/minute range is wide because model choice drives it. Pairing the platform with a $0.0032/minute model such as GPT 6 Luna or Claude 5.5 Haiku, instead of the recommended GPT 5.6 Terra at $0.064/minute, moves the platform bill more than any other single decision.
Compliance requirements: HIPAA/BAA, SSO, and custom data terms appear only under Retell’s Enterprise tier. Synthflow’s pages list SOC 2 Type 2, ISO/IEC 27001, HIPAA, and GDPR, and its contract is scoped to your needs. This is where lock-in deepens, because the compliance need and the contract arrive together.
Anti-lock-in priority: run the voice layer on a usage-based hosted platform with no annual contract, keep the glue on self-hosted n8n Community Edition, and maintain your own exports of recordings, transcripts, and booking data. Vapi’s Core tier adds 30-day retention and zero-data-retention mode; Pro extends retention to 180 days.

FAQ

These three questions cover the billing and exit details most often missed when a shop compares AI receptionist options. Each answer resolves to a published figure or a documented policy from the pricing pages linked above, not to opinion, and no hands-on testing sits behind any of them.

Do I pay for the phone number and the AI minutes separately?

Yes. The telephony layer and the platform layer bill independently. Twilio charges $1.15/month for a local number plus $0.0085/minute to receive local calls, and that bill exists whether or not an AI answers. The voice-AI platform then charges its own rate — $0.05/minute hosting on Vapi’s pay-as-you-go package, or $0.07–$0.31/minute on Retell — on top of transport.

What happens to my recordings and transcripts if I leave?

Retention is package-dependent, so check it before signing. Vapi publishes 14 days on the free package, 30 days on Core, and 180 days on Pro. Retell publishes custom retention only at Enterprise. On the glue side, n8n Starter keeps 7 days of logs and Pro keeps 30. Export your data before cancelling any layer.

Is there a free way to try this stack?

On the self-serve platforms, yes: Vapi includes $5 in free credits, Retell includes $10, and Twilio offers a $15 trial credit that covers a number and Voice testing. Synthflow is the exception — no free plan, no trial, and no published per-minute rate; every engagement starts as an annual enterprise contract.

Sources

All figures in this post come from the vendor pricing and documentation pages listed below, reviewed on 2026-10-30. Prices change; treat every number as current as of the pages reviewed and verify on the live page before committing budget. No other sources were used. Totals are arithmetic on figures from these pages, not separate quotes.

Vapi pricing — hosting rates, packages, retention windows, transport fees, and the 1,000-minute estimate.
Retell AI pricing — pay-as-you-go range, model per-minute rates, telephony rates, the estimator example, and FAQ billing terms.
Twilio US voice pricing — number and per-minute rates, current as of August 2026 per the page.
Synthflow pricing — enterprise-only packaging and compliance certifications.
n8n pricing — plan tiers, execution caps, and the definition of one execution.

ToolBrain — tool reviews, LLM comparisons, and AI workflow guides
NiteAgent — AI agent development, frameworks, and production patterns

Cross-links automatically generated from NoCode Insider.