The short version: AI phone agents can answer, book, and qualify calls for a fraction of what a receptionist costs, and for straightforward jobs (quotes, bookings, opening hours) they work well. For anything emotional, complicated, or high value, callers still hang up on them, so the real skill is knowing which calls to hand to the robot and which to keep for a human, on purpose.
Worth reading next: Why Every Small Business Website Sounds the Same Now (And It's Not Rea.
The call you never hear about
A plumber I've worked with in Kent, let's call him Dave because that's not his real name, used to lose roughly four in ten calls before 9am and after 5.30pm. Not to voicemail, not to a competitor picking up faster. They just rang out. Nobody was there. His wife did his books in the evenings and answered the phone when she remembered, which is to say, sometimes.
That's not a Dave problem. It's the default state of most one to ten person businesses in the UK. You're on a roof, in a client meeting, driving, or asleep, and the phone doesn't know that. Every ring that goes unanswered is a person who wanted to give you money and instead found the next name on Google.
So in late 2025 we put an AI voice agent on Dave's overflow line for six weeks. Not as a gimmick. As an experiment with a spreadsheet attached.
What an AI phone agent is
Strip away the marketing and it's three things bolted together: a phone number (usually through a Twilio-style connection), a speech model that turns your voice into text and back again in under a second, and a script that tells it what it's allowed to say and what it must escalate. Companies selling this kit right now include Synthflow, Retell AI, Bland, Vapi, and Goodcall, along with bigger players folding it into existing tools. Some are built for developers. Some, like Goodcall, are built so a non-technical owner can set one up in an afternoon.
The pitch is always the same: never miss a call again. What nobody puts on the landing page is what happens in the awkward middle, when the caller isn't sure what they want, or is upset, or has an accent the model wasn't trained on well. That's where the real story is.
The six week test, with actual numbers
Here's what we set up for Dave. The AI agent answered any call after four rings on the overflow line, gave opening hours, took name and postcode, asked what the job was, and either booked a callback slot in his calendar or, for anything it flagged as urgent (burst pipe, no hot water, smell of gas), texted him immediately and told the caller someone would ring within fifteen minutes.
- Missed calls before: roughly 41% of total inbound calls
- Missed calls after: 9%, mostly during the AI's own brief outages
- Calls the AI handled start to finish without a human: 58%
- Calls it correctly escalated as urgent: 14 out of 14 in the test window
- Calls where the caller hung up mid-conversation, annoyed: 11%
- Monthly cost: £89 for the platform plus roughly £40 in per-minute call charges, against the £1,400 to £1,800 a month a part time answering service would have cost
That 11% hang-up rate is the number the sales pages don't show you. Some of those were older callers who wanted a person and said so within the first ten seconds. A few were people with a complicated job who got frustrated repeating themselves when the AI misheard "the stopcock under the stairs" as something unintelligible and asked them to say it again. One was someone crying about a flood who did not want to be asked, calmly, for their postcode.
That's the uncomfortable bit most articles about this topic skip past. AI phone agents are brilliant at the calls that were never going to be hard anyway, quotes, bookings, "are you open Saturdays." They're bad, right now, at the calls where the business needed a human touch to close the sale or calm someone down. If you let the AI handle 100% of your inbound with no escape hatch to a real person, you will lose some of your best, highest emotion, highest value jobs. The trick is building the escape hatch in from day one, not bolting it on after your first bad review.
The setup that worked
What made the difference for Dave wasn't the AI itself, it was three boring decisions made before it went live.
- We wrote the escalation list first. Any call mentioning gas, flooding, "emergency," or a caller raising their voice got routed straight to a human within minutes, no exceptions.
- We let it say "I'm an AI assistant" upfront. Callers who know they're talking to a machine are more patient with it than callers who feel tricked and then figure it out.
- We reviewed the call transcripts every Friday for the first month. Not to catch the AI out, but because those transcripts are the best free market research a small business ever gets. Dave discovered three jobs a week were coming from a housing association he didn't know referred to him.
None of that needed a developer. If you can write down what your business is allowed to say on the phone, you can run this. It's the same principle behind using AI as a non-technical business owner in any other part of your business: the tool doesn't need you to code, it needs you to be specific about what "good" looks like before you switch it on.
Where the maths gets uncomfortable
Per-minute pricing is the part people underestimate. Most platforms charge somewhere between 7p and 25p a minute of call time, on top of a monthly platform fee. That sounds trivial until a caller who wants to chat, or one who keeps repeating a postcode the model keeps mishearing, runs to eight or nine minutes. Dave's longest single call in the test was nineteen minutes, a confused caller and a slightly too polite AI that didn't know how to end the conversation. That one call cost more than three normal ones.
The businesses that get burned by AI phone agents are usually the ones that treated the monthly platform fee as the whole cost. It isn't. Budget for call volume the same way you'd budget for an ad account, because it behaves the same way: unpredictable, and worth watching weekly, not annually.
Who this suits
It suits businesses with a repeatable set of questions behind most calls, tradespeople, clinics booking appointments (not diagnosing anything, just booking), salons, letting agents fielding "is the flat still available," restaurants taking bookings outside kitchen hours. Anywhere the first call is mostly logistics.
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.
It suits it less where trust is the whole sale. High end consultancy, anything where the caller is deciding whether they like you as a person before they hand over money, financial advice conversations, anything emotionally loaded. I wouldn't put an AI on the first call for my own consulting work, and I say that as someone who uses AI in most other corners of the business, from drafting to research to the kind of content experiments I've written about before, including the one where I published 569 posts in 45 days and Google only kept 5 percent of them. Volume without judgement is the pattern that keeps repeating with AI tools generally, and phones are no exception.
What it cost to get right, and what it paid back
All in, across setup, the first month of tweaking the script after listening to real calls, and the ongoing per-minute cost, Dave spent about £600 in month one and roughly £130 a month after that. In the same period he booked eleven jobs he would have lost to a ringing phone, at an average job value of around £280. The maths worked in month one and kept working. That's the same shape of recovery I've seen with other unglamorous fixes, the kind of quiet, unsexy repair that turned into the month my own website started paying me again after years of neglect. Nothing dramatic happened on the surface. The plumbing of the business, no pun intended, just started working again.
Should you build this yourself or get help
You can absolutely set up a basic version yourself in an afternoon with one of the no-code platforms. Where it goes wrong is the script writing, deciding what the AI should never say, never promise, never quote a price on, and the escalation rules, because those decisions need someone who understands both your business risk and how these models fail. If you'd rather not learn that the hard way on live customer calls, this is exactly the kind of narrow, practical project an AI implementation coach earns their fee on, a few hours of setup against months of missed calls.
And if none of the mainstream voice platforms fit your industry, it's worth a wider look, there's a longer list of lesser known but useful tools worth knowing about in 50 websites you didn't know existed but really should, several of which touch on automation and customer contact in ways most people never stumble across.
The one habit that makes or breaks it
Every business that got value from this reviewed its transcripts weekly for the first month, then monthly after. Every business that regretted it set the AI up once and never listened to a single call again. The AI doesn't improve itself. You improve it, by reading what happened and adjusting the script. Skip that and you've just replaced a missed call with a badly handled one, which is arguably worse, because now the caller thinks that's how your business runs.
Frequently asked questions
How much does an AI phone agent cost for a small business in 2026?
Expect a monthly platform fee between £30 and £150, plus per-minute call charges of roughly 7p to 25p a minute. A small trades or service business handling a few hundred calls a month should budget £80 to £250 all in, before setup work.
Will customers know they're talking to an AI?
Most will, within a sentence or two, whether you tell them or not. Telling them upfront, in a natural way, produces fewer hang-ups than letting them work it out and feel misled.
Can an AI phone agent handle bookings and payments?
Yes, most of the current platforms can check a calendar, book a slot, and send a payment link by text after the call. Taking payment details verbally, over the call itself, is where you should stop and route to a human or a secure link instead.
What's the biggest mistake small businesses make with these tools?
Switching it on and never listening to the transcripts again. The script that worked on day one won't cover every real conversation, and the businesses that keep improving it are the ones still using it a year later.
Related reading: AI Phone Answering for Small Business: The Real Costs and the Calls It Still Loses You and I Tested an AI Phone Receptionist for Three Months on a Real Small Business.
This builds on my main AI marketing guide, my main guide on the topic.
