I Let AI Answer My Business Calls for Six Weeks: What It Really Cost Me

Straight answer: an AI phone agent will pick up every call you'd otherwise miss, cost you somewhere between £80 and £400 a month depending on volume, and lose you a slice of the highest-intent callers who hang up the second they hear a synthetic voice. Nobody selling these systems will tell you that last part. I found out the hard way over six weeks on my own business line.

Why I even tried this

I was missing calls. Not loads, but enough. I run a small team, I'm often on client calls or filming, and the mobile number on my website rings through to me directly. Over one ordinary week in late 2025 I checked my call log and found nine missed calls, four of which had left no voicemail. Four people rang, got nothing, and either called someone else or gave up. That's the bit that stung. I coach businesses on exactly this kind of leak and I had one running in my own front door.

So I did what I usually do when something in my business is broken: I turned it into a six week test and wrote down every number.

What I set up

I used a conversational AI phone tool built on the newer wave of voice models rather than the older "press 1 for sales" IVR systems. There are a few players worth knowing by name if you're looking: Smith.ai and Ruby Receptionists sit at the higher-touch, more expensive end, blending AI with real humans checking in; Synthflow, Retell AI and Bland AI sit at the cheaper, build-it-yourself end, charging per minute rather than a flat monthly fee. I went with a per-minute tool because my call volume is low and unpredictable.

The setup took a genuine four hours spread across two days, not the "five minutes" the sales pages promise:

• Writing the script: who I am, what I do, what questions to ask a caller (name, company, what they need, best time to call back), and what NOT to promise (pricing, availability, guarantees)
• Recording twelve sample answers in my own voice so the AI's tone matched how I talk rather than sounding like a call centre in Manila reading from a script written by someone who's never met me
• Connecting it to my calendar so it could offer callback slots
• Testing it myself about thirty times, pretending to be an angry customer, a confused customer, and someone speaking fast with a strong accent, because that's a real caller profile and most demo videos never test for it
• Setting up an automatic email and text summary of every call to land in my inbox within ninety seconds of the call ending

Total cost for the six weeks: £1.20 per call minute at peak, averaging out to roughly £96 for the month based on 80 minutes of AI-handled calls, plus a £29 monthly platform fee. So call it £125 for month one. That's not nothing for a small business, but it's less than half of what Ruby or Smith.ai would have charged for a comparable volume with a human-blended service.

What came through the line

Over the six weeks, the AI agent handled 61 calls that I would otherwise have missed or fumbled while driving. Of those 61:

• 34 were genuine business enquiries (people wanting to know about consultancy, speaking, or training)
• 12 were suppliers, recruiters, or people trying to sell me something
• 9 hung up within the first four seconds, before the AI had even finished its opening line
• 6 were people who explicitly asked "is this a real person" and, on being told no, either hung up or asked to be called back by a human

That's the part worth sitting with. Nine hang-ups out of 61 calls is close to 15 percent, and the six who asked outright and left are another 10 percent. Roughly a quarter of the calls that reached the AI were lost specifically because it was AI. That is not a rounding error. If those had been high-value enquiries, that's real money walking away, and it's exactly the trade-off the marketing pages for these tools never mention because it makes the product look worse.

I think of a client of mine, a small heating and boiler repair firm in Leeds, who tried a similar setup a few months before I did. Their callers are often people with a burst pipe or no hot water in winter, calling in a state of mild panic. For them, the hang-up rate was worse, closer to a third, because someone stood in a cold house at 7am wants a human voice telling them help is coming, not a polite AI asking for their postcode. They dropped the AI receptionist after five weeks and went back to a human answering service, even though it cost more. Context matters more than the vendor decks admit.

The bit that changed my mind

What surprised me wasn't the AI itself, it was what the call logs revealed about my own habits. Of the 34 genuine enquiries, only 19 got a same-day human callback from me. The AI wasn't the bottleneck any more, I was. It had done its job perfectly and handed me warm leads with names, numbers, and a one-line summary of what they wanted, and I was still sitting on them for two or three days because I was busy.

That's the uncomfortable truth about most AI tools people bolt onto small businesses: the AI usually isn't the weak link. The follow-through is. I fixed mine by building a simple automation so every AI call summary also triggered a short, warm email sequence through Mailchimp, the same way I'd set up nurture sequences for years, so at least the person heard something from me within the hour even before I could call. If you've never built a follow-up sequence like that, my piece on how Mailchimp built its own audience through relentless, well-timed email is a decent template for the thinking, even applied to something as small as a missed-call follow-up.

Getting the script right matters more than the tech

The single biggest lever I pulled during the six weeks wasn't a settings change, it was rewriting the script three times. My first version sounded like every other AI receptionist: flat, over-polite, slightly robotic in its phrasing ("I would be happy to assist you with that today"). Nobody talks like that. My second version was closer to how I speak: direct, a bit blunt, no filler. Hang-up rate on the second script dropped from 19 percent to about 11 percent over the following two weeks. Same tool, same voice model, different words.

Brands that win with any channel, human or AI, tend to sound unmistakably like themselves rather than like a template. It's the same discipline behind GoPro's marketing strategy, where every piece of content, however small, still sounds like GoPro and nobody else. Your AI phone script is a brand asset whether you treat it like one or not.

Who this is wrong for

I wouldn't recommend an AI phone agent to:

• Any business where the call itself is the emergency, like plumbers, locksmiths, or anyone dealing with genuine distress on the other end of the line
• Solo service providers whose entire pitch is personal relationship and trust built over years, where a caller expects to hear your voice, not a stand-in
• Anyone who won't commit to answering the summaries within a few hours, because a fast, well-handled AI call followed by three days of silence is worse for your reputation than a simple voicemail

And I'd recommend it strongly for businesses where the call is a first enquiry rather than an emergency: consultants, coaches, agencies, B2B services, clinics booking routine appointments, tradespeople doing quotes rather than call-outs. The AI's job in those cases is triage, not comfort, and triage is exactly what it's good at.

If you want to try it yourself

A rough one-week build, based on what took me the time above:

• Day 1: Pull your last three months of call logs if you have them. Work out roughly how many minutes of calls you're missing a month. This decides which pricing tier makes sense.
• Day 2: Write the script yourself, in your own words, out loud first, then transcribe it. Don't let a template write it for you.
• Day 3: Set up the tool, connect your calendar, and record sample phrases in a voice as close to natural speech as the platform allows.
• Day 4: Test it with three people who don't know your business, ideally including someone with an accent or speech pattern different from your own, and someone who's rushed.
• Day 5: Rewrite whatever made testers hesitate or repeat themselves. That hesitation is your hang-up risk.
• Day 6: Connect the call summaries to an email or CRM workflow so leads don't sit unread.
• Day 7: Go live on a quiet day, not a Monday, so you can watch the first calls in near real time.

If your business is more complicated than a simple triage flow, or if the calls you're getting are high-value enough that getting this wrong would sting, it's worth paying for a few hours of proper setup help rather than guessing. This is the kind of small, contained project an AI consultant for small business can usually sort in a day rather than the two weeks it might take you evenings and weekends.

The numbers, laid out plainly

• Setup time: 4 hours across two days, plus 2 more hours of script rewrites in week three
• Month one cost: roughly £125 (platform fee plus per-minute usage)
• Calls handled that would otherwise have been missed: 61 over six weeks
• Genuine business enquiries recovered: 34
• Hang-up rate on first script: 19 percent
• Hang-up rate after rewriting the script: 11 percent
• Callers who explicitly asked for a human and left: 10 percent throughout
• Same-day human follow-up rate before I fixed my own process: 56 percent
• Same-day human follow-up rate after adding an automated email step: 91 percent

If you're weighing this against something simpler, remember there are lower-friction ways to capture the same intent without a voice at all. A well-placed quote or cost interactive calculator on your website catches a chunk of the same people before they ever pick up the phone, and it never gets hung up on because it's a robot.

What I'd tell you if you asked me over coffee

The tool works. That's not the interesting part. The interesting part is that it exposed how much of my "we're too busy to answer the phone" problem was a "we're not organised enough to follow up" problem wearing a missed-call costume. AI didn't fix my business. It gave me better data about what was broken, and then I had to go fix that myself. If you go into this expecting the phone agent to be the whole solution, you'll be disappointed within a fortnight. If you go into it expecting a very good triage nurse who hands you a clean list of who to call back and when, you'll probably keep it.

Frequently asked questions

How much does an AI phone agent cost for a small business?

Expect somewhere between £30 and £150 a month in platform and usage fees for a low-volume small business using a self-serve tool, or £250 to £700 a month for a blended human-plus-AI receptionist service like Smith.ai or Ruby, depending on call volume and how much of the interaction is human-handled.

Do customers hang up on AI phone agents?

Yes, and more than most sales pages suggest. In my own six week test, close to a quarter of calls that reached the AI were lost either to immediate hang-ups or to callers who asked directly for a human and left when told there wasn't one. This rate drops noticeably with a better-written script but never fully disappears.

Is an AI receptionist worth it for a tradesperson or emergency service?

Generally, no, or at least not on its own. Callers dealing with something urgent, like a burst pipe or a broken boiler, respond badly to a synthetic voice and hang-up rates tend to be higher than for routine enquiries. A human-backed answering service usually performs better for this type of business even though it costs more.

What's the biggest mistake small businesses make when setting up AI to answer calls?

Treating the AI as the finish line rather than the start of the process. The tool is good at catching calls and summarising them, but if nobody follows up on those summaries within hours, you've just built a faster way to lose leads politely instead of losing them silently.

Related reading: Why AI Pricing Advice Is Quietly Training Small Business Owners to Undercharge and Why Your AI Meeting Notetaker Might Be Breaking the Law (And Killing Your Sales Calls).

Related: ai receptionist small business test.