The published prices, and what they leave out
Every figure in this table came from the organisation’s own page, read on 4 September 2026. None of it is taken from a comparison article or another agency, because those are the two least reliable sources of pricing information available.
| Option | Published cost | What it actually is |
|---|---|---|
| Missing the call | £0, and the job | For a trade or a clinic, a missed call is usually a lost customer rather than a delayed one |
| Answering it yourself | Your time | An interruption during the work somebody is already paying you for |
| Employed receptionist | £18,000 – £22,000 a year, 38–40 hours | A person who knows the business, during their hours only |
| Human answering service | From £15 a month (messages), from £30 (appointments) | A person following your script who does not know the business |
| AI voice agent | Tens of pounds a month self-serve, or from $30,000 a year enterprise | Software that holds a conversation, books, and sends you the transcript |
Source: Salary: National Careers Service (gov.uk) receptionist job profile. Answering service: ReceptionHQ published UK pricing. Enterprise voice AI: Synthflow published pricing. All read 4 September 2026.
The number missing from every one of those “from” prices is the allowance. Not one of the answering-service tiers publishes how many calls £15 or £30 a month includes, or what a call beyond it costs — and that, rather than the headline, is what your actual bill will be made of. It is the first question to ask, and the answer should arrive in writing.
Why the market has a hole in the middle
Voice AI pricing is barbelled. At one end are self-serve tools costing tens of pounds a month, which you configure yourself and which connect to whatever they happen to support. At the other are enterprise platforms sold on annual contracts — Synthflow publishes from $30,000 a year and no monthly plan whatsoever. There is very little between the two.
That is not an accident of the market being young. A voice agent that genuinely works has to reach your calendar, your CRM and your booking system, and integration work does not scale the way software does. Self-serve products solve it by supporting a fixed list of popular tools and nothing else; enterprise products solve it with a services team and price accordingly.
Which leaves the ordinary UK small business — a dental practice, a lettings agent, a plumbing firm — choosing between a tool that will not talk to their booking system and a contract priced for a call centre. That gap is real, and it is the reason to build on the provider’s API directly rather than buy either end of it.
Working out what it is worth to you
Ignore the price list for a moment and do this instead, because it is the only calculation that matters.
- Count the calls you actually miss in a fortnight. Your phone records this; you do not have to estimate it.
- Estimate honestly what proportion of those callers rang somebody else rather than ringing back. For most trades and clinics it is the majority.
- Multiply by your average job value. That is the annual cost of the current arrangement, and it is usually a number that ends the debate.
A firm missing four calls a week, converting one in three, on a £250 average job, is losing roughly £17,000 a year to a phone nobody answers. Against that, every row in the table above is cheap — including the employed receptionist. Against a business missing one call a month on a £40 job, none of them are worth it, and the right answer is a better voicemail.
There is a market signal worth adding, because it is our own and it is unusually blunt. Advertisers pay some of the highest click prices in UK search to reach people looking for telephone answering, virtual receptionist and AI customer service help — well into three figures per click at the top end. Nobody pays that for a click unless the customer at the end of it is worth a very great deal. The people already in this market have done the calculation above, and they came out the same way.
What drives the cost of building one properly
If you are having one built rather than buying a product, the hours go somewhere specific, and knowing where lets you ask better questions of whoever quotes you.
- How many systems it must reach, and whether they have documented APIs — one is cheap, three is not, and one with no API at all is the expensive case
- What it does when it does not know: transferring to a human is simple, deciding correctly on its own is not
- Whether it writes to live records or only reads them — anything that can act unsupervised needs far more testing
- Whether your data is already clean, or spread across systems that disagree about the same customer
- How many real calls exist to test it against, because inventing test cases is slower than using yours
This site does not publish a build price for that work and does not publish other people’s either, because there is no primary source for what an integration costs and a made-up range is worse than none. What is published is the model: quoted from the hours, and then it runs on your own account with the provider at the provider’s price. Nothing is resold at a markup, and you can see the bill.
When an AI receptionist is the wrong answer
When the calls need judgement. A distressed customer, a complaint, a clinical question or a negotiation are all things a voice agent handles badly, and handling them badly in front of the customer is worse than a voicemail.
When the volume is low. Below a few missed calls a week the arithmetic above does not clear, and a diverted number or a properly-written voicemail with a callback promise does most of the job for nothing.
And when the underlying problem is elsewhere. If the phone rings constantly because the website does not say the prices, the opening hours or the service area, a receptionist — of either kind — is being paid to answer questions a page should have answered.
