The short version: An AI phone receptionist won't sound human, and customers often clock it within the first ten seconds, but it still beats the alternative because the alternative is silence. Over 90 days with a real client, an AI answering line turned 43 out of 61 previously missed calls into booked jobs. It cost £199 a month. The bit nobody tells you is that hiding the fact it's AI is the mistake, not the AI itself.
Worth reading next: The AI Chatbot Problem Nobody Warns Small Businesses About.
What an AI phone receptionist is
Strip away the marketing and it's a piece of software that answers your business phone line, talks to the caller using a voice model, checks your calendar or booking system, and either books an appointment, takes a message, or transfers the call to a human. Companies selling these include Smith.ai, Ruby, and a growing pile of smaller outfits built on top of OpenAI's voice models or similar. Prices sit roughly between £99 and £350 a month depending on call volume, with most charging a base fee plus a per-minute rate once you go over an allowance.
It is not a chatbot on your website. It answers an actual ringing phone, in something close to real time, and it has to cope with background noise, regional accents, and people who ramble.
The three months I watched one work on a real business
A client of mine runs a small boiler and heating repair firm in Kent, four engineers, no office staff. Before we set anything up, his call log showed something he hadn't quantified before: he was missing roughly a third of incoming calls during working hours because every engineer was under a boiler, not near a phone. Evenings and weekends, the miss rate was closer to 80 percent.
We put a Smith.ai style AI receptionist on the main line, plugged into his booking calendar, with a script that covered the five things customers ask: is this an emergency, what area do you cover, roughly what will it cost, can someone come today, and what's your availability. Anything outside that, it took a message and texted him a summary.
Ninety days later, the numbers looked like this:
- 61 calls that would previously have gone unanswered were picked up by the AI line
- 43 of those turned into a confirmed booking on the spot
- 12 became a callback that his team closed within 24 hours
- 6 hung up once they realised they weren't talking to a person
That's a 70 percent conversion rate on calls that used to be worth precisely nothing. A missed call is a lead sitting on the floor, in exactly the same way a website that's dropped out of Google is money you've already earned and can't collect. I wrote about watching that happen with my own traffic in the month my website started paying me again, and the maths here is the same shape: recovering something you were already losing is often cheaper and faster than trying to generate something brand new.
What it cost, in real numbers
The plan we used was £199 a month, covering up to 100 calls, with £1.20 per call after that. His actual bill across three months averaged £212. Compare that to what an answering service with real humans would cost for the same coverage, which typically starts around £300 to £500 a month for a proper 24/7 line through firms like Moneypenny or similar UK providers.
Set-up took an afternoon, not a week. That included writing the script, connecting it to his booking calendar, and testing it with ten fake calls from his own mobile to catch anything it handled badly. The main cost wasn't cash, it was the hour or two spent working out what the AI should never promise on his behalf, which for a heating engineer meant never quoting an exact price and never confirming same-day availability without checking the calendar first.
The uncomfortable bit nobody selling these tells you
Every sales page for these tools claims the voice is indistinguishable from a person. It isn't. Six callers out of 61 hung up specifically because they realised, within the first exchange, that they weren't talking to a human, and one left a one-star review mentioning it directly. That's not a bug you can train away with a better voice model, it's a small but real chunk of your callers who will always feel misled if the system pretends otherwise.
The fix wasn't a better disguise. It was the opposite: we changed the opening line from something vague to "Hi, you've reached Kent Heating Repairs, you're speaking with our AI assistant, I can book you in or take a message, what do you need?" Hang-ups dropped after that change, because the people who were bothered by AI hung up immediately and honestly, before wasting five minutes, while the people who didn't care carried on and booked the job. Being upfront about it filtered out the wrong reaction faster than pretending fixed it. Most articles selling these tools skip this because it undercuts the pitch that it feels human. It doesn't need to feel human to work. It needs to be faster than nobody answering at all.
Where it fell over
It struggled with three specific situations, consistently:
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.
- Complaints and refund requests, where callers wanted to vent, not book anything, and the AI kept trying to steer the conversation back to scheduling, which made people angrier
- Elderly callers who spoke slowly and paused mid-sentence, which the system sometimes read as the end of the call and cut in too soon
- Anything requiring judgement, like "can you squeeze me in even though you're fully booked because my boiler's leaking on the carpet right now," where the AI followed the calendar literally instead of flagging it as urgent
We solved the third one with a simple rule: any call containing the words "leak," "no heat," or "emergency" got an instant transfer to the on-call engineer's mobile, no exceptions. That single rule fixed most of the bad experiences.
Who this suits and who it doesn't
It works well for businesses where most calls are routine and bookable: trades, salons, clinics that just need appointments set, small law firms handling intake, letting agents. It works badly for anything relationship-heavy in the first conversation, like high-value B2B sales, bereavement services, or anywhere the first call is meant to build trust rather than book a slot.
If you're not technical, this is exactly the kind of AI project that's manageable without becoming a software project. I go through the same principle, of picking tools that solve one specific problem rather than trying to overhaul everything at once, in how I use AI every day without being technical. An AI receptionist is a good first project precisely because it has one job.
How to set one up without wasting a month
The whole process, done, takes an afternoon and a follow-up week of listening to real call recordings. Here's the order that worked:
- Pull your last three months of call logs from your phone provider and count how many calls you missed. Most business owners guess this number wrong by a wide margin
- Write down the five questions your customers ask most often, in their own words, not your business jargon
- Pick a provider, connect it to your existing calendar or booking tool, not a new one, so nothing has to change on your end
- Write an opening line that states plainly it's an AI assistant, then a closing rule for anything the AI can't handle: an instant transfer trigger for urgent keywords
- Run ten test calls yourself before it goes live, and listen to the first fifty real calls in week one, because that's where you catch the awkward ones
If setting up the calendar integration, the transfer rules, and the script sounds like more than you want to sort out alone, this is a small and bounded job to hand to someone who does AI implementation for a living rather than trying to learn a new platform from scratch. My AI implementation coaching covers exactly this kind of single-tool rollout, where the goal is one working system in a week, not a six-month overhaul.
Worth saying too: if you're comparing providers, don't only look at the big-name ones that show up first in search. There's a wider range of smaller, cheaper tools worth a look, and I've collected a fair few lesser-known but useful ones in 50 websites you didn't know existed, several of which are worth a scan before you commit to a monthly contract.
Frequently asked questions
Does an AI phone receptionist sound like a real person?
No, not consistently, and pretending it does causes more problems than it solves. Around one in ten callers in our test noticed within seconds and some hung up. Stating upfront that it's an AI assistant reduced complaints without reducing bookings.
How much does an AI receptionist cost for a small business?
Expect roughly £99 to £350 a month depending on call volume, usually a base fee plus per-minute charges over an included allowance. Our test client paid an average of £212 a month across three months, well below the £300 to £500 a typical human answering service charges for comparable coverage.
What kind of business should avoid an AI phone receptionist?
Anything where the first call needs to build trust or handle emotion rather than book a slot: complaint-heavy services, bereavement or crisis work, high-value B2B sales conversations. It's built for booking and answering, not for comforting or persuading.
How long does it take to set one up?
A working version can be live within an afternoon if it's connected to your existing calendar. The part that takes time is listening to the first fifty real calls in week one and fixing the two or three awkward patterns that always show up, which usually takes another few hours spread across a week.
Related reading: AI Phone Agents for Small Business: What Happens When You Let AI Answer Your Calls and I Let an AI Notetaker Sit In On My Client Calls for Three Months. Here's the Bit Nobody Warns You About.
