AI receptionist UK: what it costs, and when not to bother

An AI receptionist answers your phone when you cannot, takes a name and number, answers the handful of questions you get asked constantly, and books appointments. In the UK it costs roughly £25 to £250 a month off the shelf, or £1,500 to £3,000 to have one built around your business.

Whether you need one is a different question, and it is the one worth spending ten minutes on.

A word about every other guide on this

If you search for this, the guides you find are almost all written by companies selling an AI receptionist.

I looked at the three ranking above this one while writing it. Two of them put their own product at number one in their own comparison table. One of those discloses it, to be fair. The third is a sales page with a setup guide attached.

I do not sell an AI receptionist. I build them occasionally when somebody needs one that off the shelf tools cannot do, and rather more often I tell people they do not need one at all.

So this is the version written by somebody with nothing on the shelf.

What one actually does

Take a job I did build. A recruitment business wanted to know when companies in their sectors merged, got bought, or lost someone senior, because those are the moments a vacancy appears. It runs weekly, searches, reads, and writes the briefing.

Then it puts the briefing in a drafts folder, and a person presses send.

That is not caution on my part. The off the shelf tools will not send on your behalf without somebody authorising it. It is how the software is built.

A phone line works the same way. What an AI receptionist does, in practice, is answer at half past seven in the evening while you are having your tea.

It takes a name, a number, and roughly what the person wants. Then it answers the four questions you get asked every single day. Where are you based, do you cover my area, roughly what does it cost, how soon can somebody come out.

Then it writes it down and tells you.

It is not having a conversation. It is stopping an enquiry from evaporating.

What it costs in the UK

Most guides on this quote dollars, which is not much use in Trafford. Here are the shapes, taken from UK providers' own published pricing in September 2026. Prices move, so check before you commit to anything.

Entry level services start around £25 to £50 a month. At that price you are getting a fairly rigid script and a limited number of calls, and the cheap plans usually meter minutes in a way that gets expensive if you are busy.

The middle of the market, which is where most small businesses land, runs £100 to £250 a month depending on call volume. That buys something that books appointments, integrates with a calendar, and hands over sensibly.

A bespoke build is a different purchase. Somewhere between £1,500 and £3,000 to set up, then £100 to £250 a month to run. That only makes sense when your call handling has something genuinely specific about it, which is rarer than the people selling builds suggest.

For comparison, a human telephone answering service is roughly £200 to £450 a month, a part time receptionist £1,200 to £1,800, and a full time one comfortably north of £2,000.

The gap between £150 of software and £2,000 of person is the whole argument, and it is also why you should be sceptical of anybody presenting it as a straight swap. They do different jobs.

AI, a human service, or voicemail

Three options, and most comparisons pretend there are two.

Voicemail is free and most people under forty will not leave one. If your enquiries come from anyone who grew up with a mobile, voicemail is closer to a missed call than a message.

A human answering service is better at anything sensitive, complicated, or emotionally charged. Somebody ringing about a bereavement, a complaint, or a job that has gone wrong should get a person. It costs four to ten times more and it is worth it for those calls.

AI is better at volume, at half seven in the evening, and at the same four questions forever without getting bored.

A fair number of businesses end up running both. AI out of hours and at lunchtime, people during the working day. That is a perfectly sensible answer and almost nobody selling one option will suggest it.

Which businesses actually benefit

The pattern is consistent. High inbound call volume, simple repeated questions, and nobody sat at a desk.

Trades are the obvious one. If you are up a ladder or under a sink, you are not answering, and the person ringing has already got two other numbers off Google.

Clinics, salons, vets and physios, where a large share of calls are appointment shaped and the questions repeat.

Where it works less well: consultative businesses with low call volume and long conversations. If you get four enquiries a week and each is a forty minute discussion, an AI receptionist is a solution to a problem you do not have. I would rather say that than sell you one.

Also, and this is worth saying plainly, anywhere the first call might be an emergency. If somebody could ring you with a burst pipe or a medical worry, the escalation path has to be designed properly before anything goes live.

The bit that decides whether it works

The interesting design question is not what it can say. It is where it stops.

Somebody rings up annoyed. A job went wrong, an invoice is wrong, or they have been waiting three days for a call back. You do not want that handled well by software. You want it handled by the person who can fix it, and you want to know within about ninety seconds.

So the useful version has a short list of things it handles and a much clearer rule about what it hands over immediately. Complaints. Anything about money already in dispute. Anything where the caller uses the word urgent.

That handover rule is the whole product. Everything else is answering the same four questions politely, forever.

GDPR, recording and telling people

This is the part the buyer's guides skate over, and it is the part that is actually your responsibility rather than the vendor's.

If calls are recorded or transcribed, you need a lawful basis for it, an announcement at the start of the call, and a retention period that somebody has actually decided rather than left on the default.

You need a data processing agreement with the provider. Ask where recordings are stored. Several popular tools host in the United States, which is a decision to make deliberately rather than find out about afterwards.

And tell people it is an AI. Say it in the greeting. A caller who works it out after two minutes feels handled. A caller who was told at the start usually does not mind at all.

The Federation of Small Businesses found 92 per cent of small business owners now report worries about AI related risks, up from 73 per cent two years earlier. Most of that worry is unspecific. Being straight with your own customers is the cheapest way to stay on the right side of it.

Setting one up, honestly

Off the shelf, this is a few days rather than a project.

You write down what it should know. Opening hours, areas covered, rough prices, the four questions. That is the bit that takes longest and it is the bit only you can do.

You decide the handover rules. Then you connect a calendar if you want booking, and you point your number at it or port the number properly.

Then you test it by ringing it yourself, repeatedly, being deliberately awkward. Get somebody who does not work for you to ring it too, because you already know what it is supposed to do and they do not.

Give it a fortnight before you judge it, and read every transcript in that fortnight. That last part is not optional.

It will get things wrong in the first weeks, in the same way a new starter would. Unlike a new starter, it does not learn from being corrected unless somebody goes and changes what it was told.

How to tell whether you need one at all

None of the guides ranking for this will tell you how to decide against buying, so here it is.

Get your call log out. Most phone systems keep one, and if yours does not, your mobile does.

Count the missed calls outside your working hours over the last fortnight. Then count the ones during working hours where nobody picked up because everybody was out on a job.

That second number is usually the one that surprises people.

Then work out what one job is worth to you. If you are missing four calls a fortnight and a job is worth £400, the arithmetic makes itself. If you are missing two a month, buy nothing.

That exercise takes about twenty minutes and costs nothing.

Why it helps the quiet end more than the busy end

There is a well known study by Brynjolfsson, Li and Raymond on customer support agents, published in the Quarterly Journal of Economics. Average productivity went up 14 per cent. For the least experienced agents it was 34 per cent. For the most experienced, close to nothing.

It was a single firm and a single job type, on tooling two model generations old, so take the shape rather than the numbers.

The shape is the useful bit, and it is the opposite of how this normally gets sold. The value did not go to the best person. It went to the newest one, who now had a decent answer available at the moment they needed it.

The same applies to a phone line. An AI receptionist does not improve the calls you already handle well. It improves the ones nobody answered.

The honest limit

This does not fix a business that is not ringing anybody back.

If enquiries come in and sit for three days, the constraint is not answering the phone. It is what happens next. Putting something clever on the front of a process that already has a queue in it just gets you a longer queue, arriving faster.

It also will not sound like you. It will sound like a competent stranger who has read your notes. For a lot of trades that is a clear improvement on a voicemail nobody checks. For a business where people ring up specifically to speak to you, it is a downgrade, and you should know which one you are before you buy anything.