Straight answer: if you’ve installed an AI phone agent or chat receptionist for your small business, ring it yourself pretending to be a confused, slightly annoyed customer before you let real customers near it. Most owners never do this, which is exactly why AI receptionists are quietly costing small businesses bookings, not saving them.
Worth reading next: Should You Hire an AI Consultant? What They Do (And When It’s a Waste .
Nobody tests the thing they just bought
Over the past year I’ve watched small business owners install AI phone agents the way people used to install answerphones: switch it on, feel relieved, walk away. A plumber sets up an AI receptionist to catch calls after 6pm. A dental practice adds an AI chatbot to handle booking questions. A hairdresser’s salon puts an AI voice agent on the main line so nobody misses a call during a cut. All sensible reasons. Almost nobody then rings their own business and tries to break it.
I did this for a client last October, a heating engineer in the Midlands with three vans and a booking problem, too many calls, not enough hands to answer them. He’d set up an AI phone agent through one of the popular voice AI platforms, the kind that costs around £40 to £120 a month depending on call volume, plus 10 to 20 pence a minute over a certain threshold. It sounded fine on the demo. So I rang it myself, three times, pretending to be three different customers.
What an AI receptionist costs, in real numbers
Before the story, the money, because nobody talks about it. A basic AI phone agent for a small business typically runs:
- £25 to £60 a month for a starter tier covering a few hundred minutes
- £100 to £300 a month for a business tier with more minutes, calendar booking, and CRM sync
- 10p to 25p a minute in overage once you go past your included allowance
- A one-off setup fee, anywhere from free to £500, if you pay someone to write the call flow
That last line matters more than any of the others. The monthly fee is the cheap bit. The expensive bit is what happens when nobody writes the call flow and the AI just guesses.
The seven-question mystery call test
Here’s the test I now run on every client’s AI phone or chat agent before it goes live, and the one I’d want you to run this week if you already have one switched on.
- Ask it something slightly off-script, not “what are your opening hours” but “can you fit me in tomorrow if it’s an emergency”
- Give it a postcode or address that’s on the edge of your service area and see if it knows the boundary
- Ask about a price, even a rough one, and see whether it invents a number or correctly says it can’t quote
- Interrupt it mid-sentence, the way real customers do, and see if it copes or restarts
- Ask it to cancel or change a booking that doesn’t exist and watch what it does with that confusion
- Speak with a strong regional accent or mumble a bit, on purpose, and see how many times it says “sorry, could you repeat that”
- Ask it something totally unrelated, like whether it does weekend work when you only do Monday to Friday, and see if it makes something up to sound helpful
Run all seven, write down what it got wrong, then fix the script before a single real customer hits those gaps.
What I found when I ran it
Back to the heating engineer. Question two killed it. I gave the AI receptionist a postcode about four miles outside his actual coverage area, close enough that a human receptionist would say “let me check with him” and instead the AI cheerfully booked an appointment slot for a job he had no intention of driving to. Question three was worse in a quieter way: I asked roughly how much a boiler service costs and it gave me a specific figure, £89, that wasn’t his price at all, it was a generic average the model had clearly picked up from training data or a badly written prompt. Two real customers had already been quoted that wrong number before we caught it. He had to ring both back and explain. One booked anyway. One didn’t.
None of this was the AI being broken. It was doing exactly what it was told, which was be helpful and sound confident. Nobody had told it the boundaries of what it knew.
The uncomfortable bit most AI advice skips
Here’s the part that doesn’t get said enough: an AI receptionist doesn’t reduce your workload the way the sales page implies, it just moves the work from “answering the phone” to “auditing what the phone said.” If you don’t do that second job, you’re not saving time, you’re outsourcing your first impression to a system that will guess confidently rather than admit it doesn’t know. A human receptionist who’s unsure says “let me check.” Most AI phone agents, unless you’ve explicitly trained them not to, will produce a plausible-sounding answer instead of an honest gap. That’s the trade you’re making when you switch it on, and almost nobody tells small business owners that upfront because it makes the tool sound less magic.
This isn’t a reason to avoid AI receptionists. It’s a reason to treat the first month exactly like probation for a new member of staff, which is roughly how I now pitch it to clients when I’m brought in through hands-on AI implementation work rather than just handed a tool and left to it.
Where they earn their keep
None of this means bin the idea. Once the script is tightened, AI phone and chat agents do three things well for small businesses:
- They catch the calls that would otherwise go to voicemail and never get returned, which for most small service businesses is 20 to 40 percent of after-hours calls, based on the call logs I’ve seen from clients
- They ask consistent qualifying questions every single time, no forgetting to ask about access, parking, or the right contact number, the small details a rushed human on a Friday afternoon skips
- They hand you a clean transcript instead of a scribbled Post-it note, which means your follow-up email or text can go out same day instead of two days later when someone finally deciphers the handwriting
That third point is where the real money sits, and it’s the bit almost nobody builds.
The follow-up nobody sets up
An AI receptionist that takes a message and does nothing else is only half the system. The other half is what happens to that lead five minutes later, and this is where I see the most waste. A missed call gets logged, the AI sends a polite text, and then nothing, because nobody connected it to an actual follow-up sequence. If you’re already running email through something like Mailchimp for the rest of your marketing, the same logic that makes a good email nurture sequence work applies here: the AI call transcript should trigger an immediate, specific follow-up, not a generic “thanks for calling” template. A lead who asked about emergency boiler repair needs a different next message than one who asked about an annual service plan, and most AI receptionist setups treat both the same because nobody bothered to split them.
The businesses that get this right treat the AI receptionist as the front door and the follow-up sequence as the actual sales conversation, the same way Airbnb treats a booking enquiry as the start of a relationship rather than the end of a transaction, something worth studying in how Airbnb built trust into every stage of a booking rather than just the first click.
Chat agents have the same problem, with an extra layer
If your small business uses an AI chatbot instead of, or alongside, a phone agent, the mystery call test still applies, just typed instead of spoken. Type a slightly odd question. Type a postcode. Ask about a price. Ask something outside your actual hours or services. The failure mode is identical: confident, wrong answers, because the chatbot was set up once and never audited again. I’d also add one extra check for chat specifically: ask it the same question three times in slightly different words and see if it gives you three different answers. If it does, your knowledge base is thinner than you think, and that’s worth fixing before you point more traffic at it, the same way a good interactive tool needs testing from a real visitor’s point of view before you trust it to convert anyone.
Should you hire someone to build this
You don’t need an agency for this, but you do need someone, even a few hours of someone’s time, who will sit down and write the actual boundaries into the script: what areas you cover, what you will and won’t quote over the phone, what happens when someone asks for something you don’t do. That’s an afternoon’s work done, not a six-week project. I’ve built these call flows for clients in a single sitting once I have their actual service area, actual pricing rules, and a list of the ten most common questions their real receptionist used to field. If you’re weighing up whether to pay for that hour of setup versus letting the default script run, it’s worth reading through what an AI consultant for a small business does before you decide, because the cost of getting it wrong, the way that heating engineer found out, is usually higher than the cost of getting it built the first time.
The bit about relationships that AI can’t shortcut
I keep coming back to something I took from studying Joe Pulizzi’s approach to content and audience building, which has nothing to do with phone systems on the surface but everything to do with them underneath: the businesses that win long term are the ones that treat every touchpoint, including a missed call at 7pm, as part of a relationship rather than a transaction to be automated away. There’s a lesson in how Pulizzi built trust before he built revenue that applies directly here. An AI receptionist that sounds warm and gets the postcode right builds trust in the first thirty seconds. One that confidently gives a wrong price does the opposite, and it does it at scale, every single call, until someone bothers to check.
What to do this week
If you’ve already got an AI phone or chat agent running, block out twenty minutes and run the seven-question test yourself today. Write down every wrong answer. Fix the script or the knowledge base, not next quarter, this week. If you’re about to install one, insist on running the test before it goes live, not after the first complaint. It’s the cheapest audit you’ll ever run on your own business, and it’s the one almost nobody bothers to do.
Frequently asked questions
How much does an AI receptionist cost for a small business?
Expect £25 to £60 a month for a basic tier handling a few hundred minutes, £100 to £300 a month for a fuller business tier with calendar booking and CRM sync, plus 10p to 25p a minute in overage. Setup done, with your real pricing rules and service boundaries built in, can add anywhere from nothing to around £500 depending on complexity.
Can an AI receptionist replace a human answering the phone?
It can catch calls a human would have missed, which for most small service businesses is 20 to 40 percent of after-hours enquiries. It’s weaker at judgement calls a human makes instinctively, like whether a job is worth the drive or whether a price should be given at all. Treat it as covering the gaps, not replacing the person who knows the business.
What’s the biggest mistake small businesses make with AI phone agents?
Switching it on and never testing it themselves. Owners assume the demo call means it works in every real scenario, then find out three weeks later, usually from a customer complaint, that it’s been quoting the wrong prices or booking jobs outside the service area.
Should I use an AI chatbot or an AI phone agent for my small business?
Most small businesses that take a lot of calls, trades, salons, clinics without medical booking systems, benefit more from a phone agent because that’s where their customers already are. Businesses with younger audiences or more browsing-before-buying behaviour, retail, bookings, services people research online first, tend to get more use from a chatbot. Many now run both, provided both get the same testing.
Related reading: AI Receptionists for Small Business: What They Really Cost (And When They’re a Waste of Money) and AI Voice Agents for Small Business: What They Really Cost (And When They Fall Flat).
This builds on my main AI marketing guide, my main guide on the topic.
