How much does an AI receptionist actually cost in 2026? Real numbers, no agency-speak
We dug into Vapi, Retell, Bland, Synthflow and the custom-build path. Here's what each option costs end-to-end — setup, per-minute, hidden fees, and where the real ROI lives.
Pricing for AI receptionists in 2026 lands anywhere from £0.04 a minute to £4,000 setup + monthly. Here's the unvarnished breakdown across every popular platform plus the custom-build path — including the hidden costs nobody talks about.
An "AI receptionist" used to mean a glorified IVR menu reading from a script. In 2026, it means a voice agent that answers your phone in a conversational, human-quality voice, qualifies the caller, books into your calendar, writes the lead to your CRM, and routes anything tricky to a real person — all in 90 seconds, 24/7, for somewhere between £0.04 and £0.12 a minute.
But the headline numbers lie. The real cost depends on whether you go off-the-shelf (Vapi, Retell, Synthflow, Bland) or commission a custom build, how many calls you handle, what model you choose, and how much logic the agent needs to know. We've shipped both — including for clients who'd been quoted £4k/month elsewhere — so this guide pulls back the curtain on what each path actually costs.
TL;DR · pricing at a glance
If you only have 90 seconds, here's the spine of the guide. Real-world prices for a small business handling around 1,000 inbound calls a month with an average call length of 90 seconds.
| Option | Setup | Monthly | Per minute | Best for |
|---|---|---|---|---|
| Synthflow / Bland (template-led) | £0–£200 | £60–£200 | £0.05–£0.10 | Solo operators · 1 simple intent |
| Vapi or Retell (DIY platform) | £0–£500 | £40–£150 (model usage) | £0.04–£0.08 | Tinkerers, agencies, small SaaS |
| Off-the-shelf "agency-as-a-service" | £500–£2,000 | £200–£800 | rolled in | Owners who want it done-for-them |
| Custom build (LGP.dev path) | £1,500–£3,500 | £40–£120 + minutes | £0.04–£0.06 | Brands, multiple intents, integrations |
| "Enterprise" agency contracts | £2,500–£10,000 | £500–£3,000+ | rolled in | Mid-market with deep pockets |
The single most useful number in that table is per-minute cost, because at any meaningful call volume that's where 80% of your spend ends up. Anything in the £0.04–£0.08 band is fair value in 2026; anything above £0.10 is the platform taking a margin you could keep.
The three pricing models you'll encounter
Vendors love to obscure their pricing model behind "talk to sales" CTAs. There are really only three, and once you know which one you're looking at, the comparison gets simple.
1. Pay-per-minute (the cleanest)
You're billed for actual call duration. Most platforms charge between £0.04 and £0.10 per minute, depending on the voice model and the underlying language model. This includes inference (the AI thinking), text-to-speech, speech-to-text, and the platform's margin.
2. Per-call or per-conversation
Some platforms (Synthflow, Bland for some plans) bundle call cost into a flat per-conversation price — usually £0.20–£0.60. Sounds simple until you realise a 30-second "wrong number" costs the same as a four-minute booking. Punishes high-volume use cases.
3. Flat monthly retainer (the agency model)
Done-for-you agencies bundle setup, hosting and a usage allowance into a flat monthly fee — typically £300–£800 for SMBs, £1,500–£3,000+ for mid-market. Convenient if you don't want to think about per-minute math, but you'll pay 2–4× the underlying platform cost. The margin is the agency.
Off-the-shelf platforms compared
Four platforms dominate the developer-friendly end of the AI voice agent market in April 2026. They all do roughly the same thing — pair a phone number with an LLM and a voice model — but they differ on price, polish, and how much hand-holding they expect you to do.
Vapi
Developer-favourite, model-agnostic, BYOK supported. You pay roughly $0.05/min plus your underlying model cost (OpenAI, Anthropic, Google) and any premium voice provider (ElevenLabs etc.). All-in usually £0.06–£0.10 per minute depending on choices. Fast to prototype, requires actual coding to deploy seriously.
Retell
Closest competitor to Vapi. Slightly more polished UI, similar pricing. Their hosted voices add up at scale; BYOK to ElevenLabs/Cartesia gets you back to the £0.05/min zone.
Synthflow
No-code agent builder. Per-conversation pricing on lower tiers, per-minute on higher. Quickest path from "idea" to "working bot" but you pay for the convenience — usually 30–60% more per call than Vapi/Retell at scale.
Bland
Pitches itself as "call infrastructure for builders". Cheaper voices, decent latency, growing fast. Good middle-ground between Vapi (developer) and Synthflow (no-code). Pricing typically £0.05–£0.07 per minute.
The custom-build path
A custom AI receptionist isn't a different platform — it's the same underlying platform (almost always Vapi or Retell), but with a developer wiring the agent into your real systems. Your CRM, your calendar, your existing phone setup, your tone of voice, your edge cases.
Setup typically lands £1,500–£3,500 for a single-business agent with 3–6 intents (book a call, get a quote, reschedule, leave a message, route to mobile, FAQ). Bigger builds — multi-language, multi-intent, integrated to specific industry tools — push toward £4,000–£8,000.
After setup, you pay the platform per-minute fee (usually £0.04–£0.06 if BYOK on the LLM and voice), plus a small monthly hosting/maintenance retainer if you want one. That's it. No agency markup. No "per call" gotchas.
See a real custom build£600 all-in for the Taylor Landscaping AI receptionist + chatbot + quote engineReal-world math: 1,000 calls a month
Let's price-out a realistic SMB scenario. A trade business taking 1,000 inbound calls a month, average call length 90 seconds. Total: 1,500 minutes of agent time monthly.
| Path | 12-month total | Year 2 total | Notes |
|---|---|---|---|
| Synthflow (no-code, mid tier) | £2,160 | £2,160 | ~£180/mo all-in |
| Vapi/Retell DIY (BYOK) | £900 | £900 | £75/mo platform + minutes, no setup if you DIY |
| Custom build (LGP.dev) | £3,300 | £900 | £2,400 setup year 1 + £75/mo ongoing |
| Done-for-you agency | £6,000 | £6,000 | £500/mo flat |
| Enterprise agency | £18,000+ | £18,000+ | £1,500/mo flat |
The custom build looks expensive in year one and cheapest in year two — because the setup cost is one-off and the ongoing cost is just platform fees. The done-for-you agency keeps charging you the same £500 every month, forever, because their model is rent-not-own.
"Every agency contract I've audited for clients has had the same pattern: a £400–£800 monthly fee, a 3-year tail, and £0 of underlying platform cost in year three because the agency upgraded to a cheaper provider and pocketed the difference."
— Lewis Parker, LGP.dev
The hidden costs nobody mentions
Sticker prices are misleading. Here's where the real money goes after month two.
- Phone number rental (£1–£3/month per number, more for vanity numbers or specific country routing).
- Voice provider premium (ElevenLabs Turbo at scale runs £0.10–£0.18/1,000 chars — easy to add £30–£80/month).
- Transcription overruns (some providers charge separately for STT — Whisper-tier costs ~£0.006 per minute).
- Outbound calls if you also do follow-up campaigns (often 2× the inbound rate due to telecom regulation).
- Escalation handoff to a human (rerouting to a mobile after an AI fail = a chargeable second leg).
- CRM integration fees if your CRM nickel-and-dimes API access (HubSpot, Salesforce — looking at you).
- Recording storage if you keep transcripts for compliance (£0.02–£0.05 per call indefinitely).
When each option actually wins
Off-the-shelf platforms (Vapi/Retell DIY)
Best when: you have at least one technical person on the team, your use case is genuinely simple (1–2 intents), and you want the cheapest possible per-minute cost. Worst when: the agent needs to do anything beyond "answer + qualify + book" — you'll spend £2,000 of your own time wiring up integrations.
No-code platforms (Synthflow/Bland)
Best when: you're a non-technical owner-operator who just needs the phone answered. Worst when: you have specific brand voice requirements, custom integrations, or you'll outgrow the templated logic in six months and have to rebuild from scratch.
Done-for-you agency
Best when: you want zero involvement, can comfortably afford £500+/month indefinitely, and treat it like another bill. Worst when: you actually do the math over a 24-month horizon and realise you've spent £12,000 on a system that genuinely cost £3,000 to build.
Custom build
Best when: you want it shaped to your business (CRM, calendar, tone, multi-intent), you'd rather own the IP than rent it, and you're around long enough to recoup the setup cost (most clients break even in 6–10 months). Worst when: you genuinely have one simple intent and you don't care about ownership — Synthflow will get you there in an afternoon.
Frequently asked questions
What's the cheapest way to get an AI receptionist working today?
Vapi or Retell DIY using OpenAI BYOK and Cartesia voices. You can be live for under £30 if you can write the agent prompt yourself and skip CRM/calendar integration. Expect £0.04–£0.05/min ongoing. Anything that promises cheaper is either a free trial or a downgraded model.
Will an AI receptionist actually sound human in 2026?
Yes — at the £0.06+/min tier. ElevenLabs Turbo, Cartesia Sonic, and Vapi's native voices are now indistinguishable from a human in 90% of conversations. The remaining 10% is laughter, interruptions, and emotional context — areas where the gap will close further by mid-2026.
How long does it take to set up an AI receptionist?
No-code platforms: 2–6 hours for the basic setup, plus a week of testing and tweaking to make it sound right. Custom build: 3–7 days for a single-business agent with 3–5 intents and one or two integrations. Anyone quoting "instant" is selling you a template.
Can I have multiple AI agents on different numbers?
Yes. Most platforms let you spin up agents per phone number with separate prompts, voices and integrations — great for multi-location businesses or franchises. Pricing is typically per-minute, not per-agent, so additional agents don't multiply your monthly cost.
What happens when the AI doesn't understand the caller?
A well-built agent has explicit fallback rules: if confidence drops, it apologises, captures the caller's name and number, and either routes to a human or sends a callback request. This logic is the single most important part of the build — and what separates a £600 custom job from a £6,000 one.
Lewis runs LGP.dev — bespoke software for businesses, from Newark, UK. He's built AI agents, multi-tenant SaaS, charity platforms and trade-business tooling for clients across the UK.
Got an idea? Talk it through →
Got a build like this in your head?
Free 30-minute call. Fixed quote in 48 hours. Source code yours. If we're not the right fit, I'll say so up front.