Affiliate disclosure: This article contains affiliate links for Retell AI and n8n. If you sign up through them, Automated Wealth Lab may earn a commission at no extra cost to you. The pricing math below is built from published list prices, not from any vendor’s sales deck.
The product nobody wants to build — and every plumber needs
Voice AI has a PR problem: everyone pictures call centers. The actual market is the opposite end of the size spectrum. A two-location dental clinic, a plumbing company with four techs, a law firm with one paralegal answering the phone — these businesses lose real jobs to voicemail every week, and they can’t justify a $3,000/month human answering service.
The model: white-label a voice AI receptionist, sell it as a monthly retainer per business, and keep the spread between price and per-minute cost. No call center, no seats — recurring revenue on per-minute infrastructure.
The business model: retainer math, not seat math
You are not selling software seats. You are selling an answered phone. The pricing shape that works: a flat monthly retainer per business, positioned against the human alternative (~$250–350/mo for a basic human answering service covering limited minutes).
Retainer examples positioned against that benchmark:
| Your retainer | Target client | What you promise |
|---|---|---|
| $197/mo | Solo operator, low call volume | 24/7 answering, message capture, SMS summary |
| $297/mo | Busy local service (plumber, clinic) | + appointment booking, call routing, CRM notes |
| $497/mo | Multi-location or high-volume | + priority support, monthly call analytics, custom greetings |
The margin table
The core question: what do the minutes actually cost you? The realistic config lands around $0.10–0.13/minute all-in ($0.055 voice infrastructure + $0.015 standard voice + ~$0.04 mid-tier model + $0.015 US telephony).
| Client profile | Minutes / mo | Your infra cost (at ~$0.12/min + $2 number) | You charge | Gross margin |
|---|---|---|---|---|
| Quiet solo (30 calls Ă— 2.5 min) | ~75 | ~$11 | $197 | ~$186 (94%) |
| Typical local service (120 calls Ă— 2.5 min) | ~300 | ~$38 | $297 | ~$259 (87%) |
| Busy clinic (300 calls Ă— 3 min) | ~900 | ~$110 | $497 | ~$387 (78%) |
Two honest notes: per-minute billing covers the entire call duration including silence — budget for billed minutes, not talk minutes. Concurrency is generous (20 simultaneous calls included before $8/month per extra slot); each client gets a dedicated phone number ($2/month).
Cost-per-minute breakdown across platforms
All figures from published pricing pages or verified October 2026 breakdowns. Note that voice AI platforms quote in layers — the headline rate is never the whole call.
| Platform | Headline rate | Real per-minute, realistic config | Platform fee | Best fit |
|---|---|---|---|---|
| Retell AI | $0.07–$0.31/min blended | ~$0.10–0.13/min (platform voice + mid model + US telephony) | None (pay-as-you-go) | Developer-built agents you resell; fully computable cost stack |
| Vapi | $0.05/min hosting | ~$0.08–0.13/min (hosting + provider pass-through) | $0 (Build) or $29/mo (Core) | You want provider-level control and pass-through pricing |
| Goodcall | $79/mo flat (entry) | Flat — unlimited minutes, capped by unique customers (100) | Built in | Turnkey, zero-build; you resell at markup |
| Dialzara | $29/mo (60 min) → overage $0.48/min | ~$0.48/min overage is the killer | Built in | Entry-level demo, not client scale |
| Rosie | $49/mo + $0.25/min | Scales fast with volume | Built in | Trades-specific, quick setup |
The pattern: turnkey tools (Goodcall) trade margin for simplicity; infrastructure platforms (Retell, Vapi) trade simplicity for margin. Past a handful of clients, the infrastructure route wins — Goodcall’s $79/mo entry is competitive for a first client, but then you’re reselling a fixed product with thin markup.
Retell’s stack is the most transparent: every layer has a published number (voice infra $0.055, platform voices $0.015, ElevenLabs $0.040–0.10, LLMs $0.0016–$0.32/min, US telephony $0.015, $2/mo per number). The advertised $0.07 floor needs the cheapest model and your own SIP trunking — plan margins on $0.10–0.13. Also: prompts over ~4,000 LLM tokens can scale the billed duration of the whole call, so keep prompts lean.
The n8n wiring
The voice agent handles the call. n8n handles everything around it:
- Retell post-call webhook → n8n Webhook node. Retell fires a webhook with the transcript, summary, duration, and disposition; n8n’s Webhook trigger receives it as JSON.
- Parse and route. A Code or Filter node extracts the outcome: appointment booked, message taken, spam, or needs-human-callback.
- Push to the client’s systems. Write the summary to the client’s CRM (HubSpot, GoHighLevel), append a row to a shared Google Sheet, create a calendar event for bookings.
- Notify the business owner. An SMS to the owner with the call summary — this is the feature clients actually feel.
- Log costs. A parallel branch logs minutes and estimated cost per client to your own sheet — your margin dashboard.
Wiring cost: trivial — one n8n execution per call, about a cent on Cloud Starter, n8n self-hosting for a tiny droplet. The orchestration layer adds single-digit dollars per client per month.
The compliance note (don’t skip this)
This is the section that separates a real business from a weekend project:
- Call recording consent is jurisdiction-specific. In the US, some states require one-party consent while others require all-party consent. Your agent’s opening greeting should announce the call may be recorded — that single sentence solves most of the problem in most jurisdictions. Retell also exposes opt-out recording and transcription controls.
- GDPR applies if you serve EU clients. You need a data processing agreement (DPA) with your vendors, defined data retention, and the ability to delete call data on request.
- Outbound is a different legal animal. If you add outbound (reminders, follow-ups), the TCPA in the US and equivalent regulations elsewhere impose far stricter consent and opt-out requirements. Keep your v1 product inbound-only.
- Regulated verticals need more. HIPAA compliance is available on Retell enterprise contracts with a BAA — budget for it before you quote a clinic, not as a retrofit. Vapi lists a $2,000/month HIPAA add-on.
Not legal advice — a checklist of the questions a serious buyer will ask. “I don’t know” is not a viable answer when you’re selling into their phone system.
Honest caveats before you sell client #1
- Support is the real product. Local business owners will call you when the agent mispronounces the street name, books the wrong slot, or sounds “too robotic.” Price your retainer to cover 2–3 support touches a month per client, or this becomes a job.
- Voice quality is a demo-killer. The difference between a $0.015/min platform voice and a $0.04–0.10/min ElevenLabs voice is audible on a real phone line. Test on an actual call, not laptop speakers, before committing to the cheap voice.
- Churn looks like “we hired a receptionist.” Clients outgrow you when they can afford a human. That’s fine — it means your pricing ceiling is real, so keep acquisition cheap (local SEO, partnerships with web designers who serve the same verticals).
- Start with one vertical. Home services have predictable call patterns; clinics and legal carry compliance overhead; restaurants need reservation integrations. Depth beats breadth at this scale.
FAQ
How much does it cost to run an AI receptionist per local business?
At ~$0.12/min all-in, a typical local service (~300 min/mo) costs ~$38 in infrastructure + $2 for the number, plus a few dollars of n8n orchestration. A quiet solo (~75 min) costs ~$11. Published-rate computations, checked October 2026.
Retell vs Vapi — which is cheaper for a receptionist micro-SaaS?
Realistic all-in costs overlap ($0.08–0.13/min on both). Retell bundles everything into one bill with published per-layer rates; Vapi charges $0.05/min hosting and passes provider costs through at their own prices. Pick on control and developer experience, not on the headline number — the headline is never the whole call.
Can I just resell Goodcall instead of building?
Yes — Goodcall’s $79/mo entry tier (unlimited minutes, 100 unique customers) is the lowest-friction path. The tradeoff is margin and differentiation: you’re reselling a fixed product, so your markup is thin and your offering is identical to anyone else’s Goodcall resale. Building on Retell/Vapi costs more setup time but gives you 80%+ margins and a product you control.
Do I need n8n for an AI receptionist?
Not for the call itself — Retell handles that. You need n8n (or equivalent) for everything around the call: pushing summaries to the client’s CRM, booking calendar events, texting the owner, and logging your per-client costs. Webhook-triggered, it’s roughly a cent per call on n8n Cloud, near-zero self-hosted.
Is it legal to record AI-handled calls?
It depends on jurisdiction — US states split between one-party and all-party consent rules, and GDPR governs EU calls. The practical move: have the agent announce that the call may be recorded in its opening greeting, document a retention/deletion policy, and keep outbound out of v1. This is a checklist, not legal advice.
What does Retell’s affiliate program pay?
Third-party affiliate directories report ~15% recurring for Retell, but sources disagree on duration (12 months vs 2 years), and we could not verify the current terms on Retell’s own pages. Verify the live terms in Retell’s affiliate dashboard before quoting any rate.
AWL verdict
The AI receptionist retainer is one of the cleanest micro-SaaS models in voice AI: ~$38 of infrastructure sold for $297/month to a local service business is an 87% gross margin with a support burden you can actually handle solo. Retell is the infrastructure pick because every cost layer is published and computable — margin math you can do before you sell. The risks are not technical: they’re recording-consent compliance, voice-quality demos, and the support load of non-technical clients. Start with one vertical, one flat retainer, and inbound calls only. The margin table above is your pricing floor; everything above it is salesmanship.
Methodology
What was verified: Retell’s per-minute cost structure was read directly from retellai.com/pricing in October 2026 ($0.07–$0.31/min blended range; $0.055/min voice infra; platform voices $0.015/min; ElevenLabs $0.040–0.10/min; LLM rates $0.0016–$0.32/min; US telephony $0.015/min; $2/month numbers; 20 free concurrent calls then $8/mo; full-call-duration billing including silence). Vapi’s $0.05/min hosting + pass-through structure, Goodcall’s $79/mo entry tier, Dialzara and Rosie pricing were cross-checked across published October 2026 breakdowns. Retell’s ~15% recurring affiliate rate appears only in third-party directories with conflicting durations — stated as unverified. What was NOT run or tested: no voice agent was built, no calls were placed or billed, no client onboarding was performed, and no compliance review was conducted with counsel. Minute volumes per client profile (75/300/900) are stated illustrative assumptions. Recording-consent and GDPR notes are general information, not legal advice.
Sources
- Retell AI pricing page — per-minute layers, billing rules, phone numbers, concurrency
- Retell AI pricing breakdown — worked 10,000-minute example
- Vapi AI per-minute pricing explainer
- AI receptionist market overview — verified vendor list prices (checked Oct 3, 2026)
- AI voice-bot competitor pricing comparison — Goodcall, Dialzara, Rosie, Smith.ai tiers
- Affiliate program directory — Retell AI ~15% recurring listing
- n8n affiliate program terms — 30% of Cloud referrals, 12 months
