The short version: AI receptionists answer calls well, sound convincing within about 20 seconds, and are cheap compared to a human. The problem isn't the voice, it's the moment a caller says something the script didn't expect, and that moment costs small businesses more money than the monthly fee ever saves them. I found this out by ringing my own business pretending to be a stranger, and I'd rather you learn it from me than from a missed sale.
Why I rang my own number as a stranger
A client asked me last month whether she should swap her part-time answering service for an AI receptionist. She runs a small events staffing company in Manchester, five staff, decent bookings, and she was paying £480 a month for a human answering service that picked up calls when she couldn't. She'd been quoted £79 a month for an AI version that promised to "never miss a call again."
Before I told her yes or no, I set one up on my own business number for a fortnight, using a real spare mobile so it felt like a genuine test, and I rang it seven times pretending to be different types of caller: a straightforward enquiry, someone with a vague question, someone trying to negotiate, someone rude, and someone who wanted to book something on the spot. I didn't tell my own team I was doing it. That felt important, because staff will always behave differently if they know they're being watched, and I wanted to hear exactly what a real stranger would hear.
What the AI receptionist got right
It was better than I expected on the basics. It picked up in two rings every time. It gave a clear, warm greeting with the business name spoken correctly, which sounds like a small thing until you remember how often human answering services mispronounce a business name they've been given in a spreadsheet. It took a message accurately, texted me a summary within about 90 seconds, and it never once sounded bored, which after 20 years of running businesses I can tell you is rarer among humans than anyone likes to admit.
It also handled the straightforward enquiry call almost perfectly. Someone asking "what are your opening hours" or "do you cover Leeds" got a fast, correct answer without being passed to voicemail. For a solo trader or a small team who currently lose calls to voicemail because they're on a job, that alone is worth the £79 a month. A missed call that goes to voicemail converts at roughly half the rate of one that's answered live, according to call-tracking data most agencies quote around 30 to 35 percent conversion for answered calls versus 10 to 15 percent for voicemail follow-ups. That gap is real money.
Where it fell over, and it cost a real booking
The negotiation call is where it broke. I called asking about staffing a corporate event for 200 people and said, "I've had a quote from someone else for £3,900, can you match that or come close?" A human receptionist, even a part-time one, would say "let me get someone to call you back within the hour" and flag it as hot. The AI receptionist said, cheerfully, that it couldn't discuss pricing and offered to book me a callback "within 2 business days." Two business days. For a live enquiry worth roughly £4,200 based on the actual booking size my client normally quotes for that headcount.
By the time a human calls back two days later, in my experience running lead follow-up for clients over the past decade, somewhere between a third and half of urgent enquiries have already booked with a competitor. That's not an AI problem exactly, it's a configuration problem, but almost nobody configures the escalation rules when they set these things up, because the sales page shows you the easy demo, not the awkward one. This is the bit that gets left out of most "we tried an AI receptionist" reviews, because most of those reviews are written by people who rang once, asked a soft question, and declared it magic.
The rude caller test
I also rang and was deliberately short and irritated, the way a annoyed customer sounds. The AI stayed pleasant, which sounds like a win, but it also didn't flag the call as a complaint in the summary it sent me. A human would have written "this one sounded really unhappy, might want to call back yourself." The AI's text just said "caller enquired about staffing services." That flattening of tone is the quiet cost nobody mentions: you get the transcript, but you lose the read on the room.
The real cost breakdown
Here's what the numbers look like once you go past the headline price:
- Human answering service (part-time, UK): £300 to £600 a month for business-hours cover, more for 24/7
- AI receptionist (Rosie, Goodcall, and similar tools): £40 to £150 a month base, often with per-minute overage once you go past 100 to 300 minutes included
- Setup time to write scripts, pricing rules, and escalation triggers: 3 to 6 hours if you do it yourself, more if you want it handling bookings and calendar syncing
- Cost of one missed high-value enquiry because escalation wasn't set up: whatever your average job or client is worth, which for my client was £4,200
On paper the AI option saves £350 to £450 a month. In practice, if it fumbles even one enquiry a quarter the size of that staffing job, the "saving" is gone and then some. This is the maths most sales pages for these tools never show you, because it's not in their interest to show it.
How to test yours before you switch
If you're weighing this up, don't take a vendor demo's word for it. Do what I did:
- Ring from a number the system won't recognise, at least five times, over different days
- Include one easy call, one vague call, one negotiation or price objection, one complaint, and one where you ask for something slightly outside what it's set up for
- Check how fast and how accurately the message reaches you, not just whether it does
- Check whether it flags urgency or tone, not just topic
- Ask it directly, "can I speak to someone now," and see what happens
If it fails the negotiation or urgency test, that's not a reason to scrap the idea, it's a reason to spend the extra hour or two building a proper escalation rule before it goes live on real customers. Most providers let you set a trigger word or phrase, like "quote" or "urgent" or "today," that routes straight to a text alert or a live transfer. Almost nobody sets these up on day one because it's not the default.
Where it fits with the bigger picture
I think about this the same way I think about businesses like Airbnb, whose entire brand strategy is built on the idea that the experience of the interaction is the product, not just the transaction itself. A phone call is often a stranger's first real contact with your business. Getting the tone right matters as much as getting the facts right, and that's the piece an AI receptionist struggles with, however good the words it says are.
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.
Ahrefs built a lean, high-output team partly by automating everything that didn't need a human judgement call, and their approach to scaling without bloating headcount is worth studying if you're deciding what to automate and what to protect. Phone answering sits right on that line. The routine 70 percent of calls, hours, location, basic availability, can go to AI without much risk. The 30 percent that involves money, complaints, or urgency needs a human, or at minimum a very well configured escalation rule.
When it's worth bringing in help to set this up right
If you're running a service business where a missed enquiry is worth hundreds or thousands of pounds, it's worth an hour of proper setup rather than accepting the defaults. This is exactly the kind of small, high-use fix I walk clients through in my AI implementation coaching, because the tool itself rarely fails, the configuration does. A consultant won't make the AI smarter, but they will make sure the £4,200 call doesn't sit in a queue for two days.
Alex Hormozi talks a lot about how the offer matters more than the traffic, and there's a version of that lesson that applies here too: the best AI receptionist in the world doesn't matter if the escalation path around your highest-value enquiries is left on the factory setting.
The uncomfortable truth
Here's the bit that made my client uncomfortable when I told her. Her part-time human answering service, the one costing £480 a month, was converting the negotiation calls better than her own sales team did, because the woman answering had picked up the rhythm of the business over three years and knew when to say "let me get her on the phone right now." Swapping to AI to save £400 a month wasn't really a technology upgrade, it was a decision to trade a skill she'd built up for years for a script that doesn't know it's losing money. She kept the human for the phones and put the AI receptionist on her website chat instead, where the stakes per conversation are lower. That's the honest trade most people don't want to make, because "AI receptionist saves you £400 a month" is a much better headline than "keep paying the human who's quietly good at her job."
If you want a version of this test for your own numbers, the same logic that works for pricing pages and quote calculators applies here. I've written before about how interactive tools change conversion behaviour, and the principle carries over to phone calls: the moment a customer has to think about money, the system needs a human-shaped answer, not a polite deflection.
Frequently asked questions
Do AI receptionists save small businesses money?
Often yes on the base monthly cost, typically £40 to £150 a month against £300 to £600 for a part-time human service, but the saving disappears fast if the AI mishandles even one high-value enquiry a quarter, so the real test is escalation quality, not price.
Can an AI receptionist handle price negotiations or quotes?
Most cannot and shouldn't try. The safer setup is to have it recognise trigger words like "quote," "price," or "urgent" and route those calls straight to a text alert or live transfer rather than let the AI attempt to negotiate.
How do I test an AI receptionist before switching my whole business over?
Ring it from an unrecognised number at least five times over different days, including one easy question, one vague question, one price objection, one complaint, and one direct request to speak to a person, then check whether the summaries you receive capture urgency and tone, not just topic.
Is it better to keep a human answering service instead of switching to AI?
If your average enquiry is worth several hundred pounds or more and your current human answering service is converting well, keep it, and use AI for lower-stakes channels like website chat or after-hours basic queries instead of replacing the phone entirely.
Related reading: AI Meeting Note-Takers: What Nobody Tells You Before You Turn One On and Instagram Marketing Strategy: How They Built a Brand That Wins.
