The short version: AI phone answering works brilliantly for booking, rescheduling, and basic questions, and it will stop you losing calls to voicemail. It works badly for anything emotional, urgent, or complicated, and most "AI receptionist" products are quieter about that than their marketing suggests. Expect to pay somewhere between £60 and £350 a month depending on call volume, and expect at least one call in the first fortnight to go wrong.
Last spring, a client of mine, a small plumbing and heating firm in Kent with three engineers and one very overworked office manager, was missing four in ten calls. Not because they were lazy. Because when you're up a ladder or under a boiler, you cannot answer a phone. Every missed call was a potential £180 callout going to a competitor instead.
She'd tried a human answering service before and it cost too much for the call volume she had. So we tried an AI phone answering system instead, for real, for 30 days, with real customers, and I want to tell you what happened rather than what the sales page promised.
What "AI answering the phone" means in 2026
There are two very different products hiding under this label, and almost nobody tells you which one they're selling you.
- Genuine AI voice agents, built on platforms like Vapi, Retell AI or Bland, using a text-to-speech voice and a language model to hold a real-time conversation. No human involved unless the call gets escalated.
- Human-plus-AI hybrids, where a real person (often based in the Philippines or South Africa) answers using an AI-generated script, AI call summaries, and AI-drafted follow-up texts, but the voice on the phone is a person. Smith.ai and Ruby Receptionists both sit closer to this end, and they're upfront about it if you ask, but the marketing copy leans heavily on the word "AI" without spelling that out.
This is the bit most articles gloss over. If you're paying £300 a month expecting a robot and you're getting a human in a call centre with a script, that's not necessarily worse, it's often better for tricky calls, but it's a completely different product and a completely different privacy conversation with your customers. Ask directly: "Is a human or a model answering this call?" before you sign anything.
What we set up
We went with a genuine AI voice agent, hooked into her existing booking calendar, with a UK-accented voice (customers hung up on the American default voice twice in testing, so that mattered more than I expected).
- Wrote a call script covering the eight most common reasons people call: booking, rescheduling, "is this an emergency," pricing questions, and complaints.
- Fed it three months of call recordings so it learned the actual phrases customers use, not the phrases we assumed they'd use.
- Set a hard rule: anything mentioning "no heating," "leak," "smell of gas" or "elderly relative" routes straight to a human within one ring, no AI conversation at all.
- Connected it to the calendar so it could book, move, or cancel appointments live on the call.
- Set up an SMS follow-up so every caller got a text confirmation, because people don't trust a booking they can't see written down.
- Tested it ourselves for two days, pretending to be difficult customers, before a single real call touched it.
Cost: £249 a month for up to 300 minutes of calls, then £0.79 a minute after that. Her call volume that month came to 340 minutes, so the real bill was £337.90. Compare that to the human answering service quote she'd had before, £520 a month minimum, and the maths made sense even before you count the calls she stopped losing.
The numbers after 30 days
- Missed call rate dropped from 41% to 6%.
- Bookings made directly by the AI without any human involvement: 58% of all calls.
- Calls escalated to a human because the AI flagged them (emergencies, complaints, anything it wasn't confident about): 22%.
- Calls where the customer hung up specifically because they realised it was AI: 4, out of roughly 190 total calls.
- Estimated revenue from bookings that would otherwise have gone to voicemail: around £2,100 for the month, against a cost of £338. That's the number that convinced her to keep it.
Where it worked better than I expected
Booking and rescheduling were nearly flawless. The AI never got tired, never sounded irritated at the fifth "can I move Tuesday to Thursday" call of the day, and it answered on the first ring every single time, which alone changed how customers talked about the business. Two people mentioned unprompted that they were surprised how quickly the phone got picked up.
It also handled pricing questions consistently, something her office manager admits she used to answer differently depending on her mood and how busy she was. The AI quoted the same callout fee every time. That consistency is worth more than it sounds, because inconsistent pricing over the phone is exactly the kind of thing that ends up as a bad review.
Where it went wrong, and the bit other articles don't say out loud
Here's the uncomfortable part. Around day nine, an elderly customer called about a boiler that had stopped working overnight, in January, and instead of hearing distress in her voice and escalating immediately, the AI worked through its script politely asking clarifying questions about the make and model of the boiler. She hung up. She called back twenty minutes later, annoyed, and got a human that time. We fixed the trigger words afterwards, but it happened, and it happened because we hadn't imagined every way a distressed person talks.
This is the truth the AI receptionist companies don't put on their landing pages: the failure mode isn't the AI sounding robotic, most of them sound fine now, thanks to better voices. The failure mode is the AI being perfectly polite and perfectly unhelpful in the ten seconds a stressed human needed to feel heard. You cannot script your way out of every version of a bad day, and no amount of training data fully closes that gap. Any business selling emergency services, urgent repairs, or anything touching someone's health or safety needs a much lower bar for handing off to a human than the default settings suggest, and most owners only discover the right settings after something goes slightly wrong first.
The other thing nobody mentions: customers increasingly can tell, and a small number actively resent it. Not because the voice is bad, but because they feel like they're being managed by a system rather than spoken to by a person, especially older customers and especially anyone calling with a complaint. Complaints are the one call type I'd now route to a human by default, always, no exceptions, because an AI apology, however well phrased, tends to make angry people angrier.
If you're thinking about trying this for your own business
Start smaller than the sales calls suggest. Most providers will let you run a trial with call forwarding only during specific hours, evenings and weekends for example, rather than switching your whole line over on day one. That's exactly how I'd recommend testing it if you're not technical and this whole area feels intimidating, because the setup itself is mostly filling in a script and recording a handful of example calls, not coding anything.
Budget realistically. For a business taking under 100 calls a month, expect £60 to £150. For 200 to 400 calls, expect £250 to £400, plus per-minute overage if you go over your allowance. Ask for the overage rate before you sign anything, because that's where the bill quietly doubles.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
Decide your escalation rules before you go live, not after. Write down every phrase that should skip the AI entirely and go straight to a human. Test it with your worst-case caller in mind, not your average one.
And if all of this sounds like more decision-making than you have time for on top of running the actual business, this is precisely the kind of narrow, practical project worth bringing in outside help for rather than guessing your way through six weeks of trial and error. I've written before about what an AI consultant costs for exactly this kind of small, contained project, and for something like a phone system it's usually a few hours of setup work, not an ongoing retainer.
What I'd tell a friend who asked
Use it for booking, rescheduling, opening hours, pricing, and anything routine and repeatable. Do not use it as your only line for anything an upset or frightened person might call about. Test it on your own business for two weeks before you trust a single real customer to it, the same way I'd tell anyone to be sceptical before trusting any AI output at scale, whether that's blog content nobody reads or a phone call nobody wanted to have with a robot. And be honest with customers if they ask. Nobody has ever thanked a business for being caught pretending a machine was a person.
Frequently asked questions
How much does an AI phone answering service cost for a small business?
Realistically between £60 and £150 a month for under 100 calls, and £250 to £400 for 200 to 400 calls, with per-minute overage charges once you exceed your plan's allowance. Always ask for the overage rate in writing before signing.
Can customers tell they're talking to AI?
Increasingly yes, especially older customers or anyone calling with a complaint. Voice quality has improved a lot, but the giveaway is usually the AI staying calm and scripted through a moment that needed empathy, not the voice itself sounding robotic.
Is an AI receptionist AI, or a human using AI tools?
It depends on the product. Some services use a voice AI end to end. Others, including several well-known "AI-powered" answering companies, use human agents supported by AI scripts and summaries. Ask directly which one you're buying before you commit.
What calls should never go to an AI receptionist?
Anything involving an emergency, a complaint, a bereavement, or clear distress in the caller's voice. Set escalation rules for these before launch, and set the trigger bar lower than feels necessary, because you will only find the right level after something almost goes wrong.
Related: looking for an AI consultant for small business? Here is how I can help.
Related reading: AI Phone Receptionists for Small Business: What They Cost and Where They Quietly Lose You Money and The AI Subscription Creep Nobody Warned Small Business Owners About.
