Straight answer: an AI phone agent will save a small business real money on the boring 80 percent of calls (opening hours, quotes, booking a callback), but it will lose you a handful of higher value calls if you do not tell people it is AI or give it a fast human escalation route. I ran one for 30 days on my own business line and the numbers below are real, not estimates.
Why I even tried this
My office manager left in September for a job nearer home, which happens, and rather than rushing to hire someone at £26,000 to £28,000 a year plus employer National Insurance and pension contributions (call it £31,000 all in, for a UK-based full timer), I decided to test something I had been putting off for two years: an AI phone agent answering my main business line for a full calendar month.
I run a small consultancy, not a call centre, so most of what comes in is predictable: people asking about speaking fees, people wanting a quote for an AI consultant for small business project, journalists chasing a comment, and the odd supplier trying to sell me something I did not ask for. I wanted to know whether a machine could handle that first filter without me losing anyone I wanted to talk to.
What I set up
I used a voice AI platform (there are several worth naming for context, Vapi, Synthflow and Retell AI are the ones I tested demos of before choosing one) connected to my existing landline number through call forwarding. No new number, no confusing my regulars. Setup took me two afternoons, not the "five minutes" the sales pages promised.
- Wrote a decision tree script covering the eight most common reasons people call me, based on 90 days of old voicemail transcripts
- Recorded a short human intro line clarifying they were speaking to an AI assistant, not me
- Set escalation triggers: the words "complaint," "urgent," "quote," or "press" transferred the call live to my mobile within four rings
- Added a compliance line up front for call recording, which matters in the UK under PECR rules, not just GDPR
- Tested it with six friends and two clients before going live, and rewrote three sections because the AI misheard "speaking fee" as "speeding" twice
The numbers after 30 days
214 calls came through the line in total. The AI handled 178 of them without any human involvement at all, mostly people wanting my rates, my availability, or a link to book a discovery call. It sent 38 people straight to a booking link and 12 of those became actual paid consultations within the month.
The other 36 calls were escalated live to me or my mobile, exactly as designed. But here is the number that stung: 6 callers hung up the moment the intro line said "AI assistant," and I know from caller ID that at least two of those were existing warm leads who never rang back. That is roughly 1 in 36 of the calls I most wanted, gone in the first eight seconds.
On cost, the AI platform ran me £180 for the month including call minutes, against the £31,000 a year (roughly £2,580 a month) a full time receptionist would have cost me. Even accounting for the lost calls, the maths is not close. But maths was never really the whole question.
What it got right
It never had a bad day. It never sounded tired at 4:45pm on a Friday. It answered on the second ring, every time, which is more consistent than most humans manage, myself included. For pure information requests, "what are your rates," "do you work with retail brands," "can I get a callback tomorrow," it was faster and more consistent than any receptionist I have ever employed.
It also caught something I did not expect: three cold sales calls from software vendors got politely filtered out entirely, saving me the fifteen minutes each of those calls usually costs me. That alone paid for a chunk of the platform fee.
Where it fell down, and the bit most people setting these up quietly hide
The uncomfortable truth is that a lot of businesses running these systems do not disclose it is AI, because disclosure costs them calls, exactly the way it cost me those 6. I chose to disclose because Ofcom guidance and basic honesty both point the same direction, and because the reputational risk of a caller finding out later that they were fobbed off by a bot, without being told, is worse for a personal brand business like mine than losing a small percentage of jumpy callers up front. If you run a business where trust is the entire product, hiding it is a short term win and a long term liability.
The AI also could not read tone. A caller who was clearly upset about a delivery delay (I do occasional physical product bundles alongside consulting) got a chirpy, unbothered response before the escalation trigger kicked in on the word "refund." A human receptionist would have picked up the frustration in the first three words and adjusted. The AI needed the trigger word. That gap, reading a person rather than parsing their sentence, is exactly where automation still loses to a person, and it is the same gap that shows up whenever a brand tries to swap warmth for efficiency without thinking it through. It is worth looking at how Calm has built its entire marketing strategy around not rushing that feeling, or how Duolingo turned automation into personality rather than a mask, because both are proof that automating the delivery does not mean automating the feeling.
The step by step, if you want to try this yourself
- Pull 60 to 90 days of your own call or voicemail history and list the actual reasons people ring you, not the reasons you assume
- Write a decision tree script for the top 6 to 10 reasons, with plain, short questions, not corporate phrasing
- Record or generate a clear disclosure line at the very start of the call
- Set 3 to 5 escalation trigger words that immediately route to a live person, and test them out loud, not just on paper
- Forward your existing number rather than issuing a new one, so nobody has to relearn how to reach you
- Run a two day live test with people who already know you, before going live with strangers
- Check your weekly call logs for missed intent, mine flagged "speeding" instead of "speaking fee" in week one, an easy fix once I saw it
If none of that sounds like something you want to build yourself, this is the exact kind of project an AI implementation coach can set up with you in a single working session rather than the two afternoons it took me solo.
Who should do this, and who should not bother
If your calls are mostly information requests, availability checks, or booking requests, an AI phone agent will pay for itself inside the first month and free up hours you were spending on repetitive answers. Solo consultants, home-based service businesses, tradespeople who miss calls while on a job, and small clinics booking appointments (not giving medical advice, just booking) all fit this well.
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.
If your business runs on relationship-led, high-ticket sales where the first call is where trust starts, be careful. The same instinct that makes GoPro's marketing strategy work, built entirely on real people telling real stories rather than polished automation, applies here too. Automating the front door of a trust-based business can quietly cost you the exact leads you built the business around.
A smaller, cheaper first step for a lot of businesses is not a phone agent at all, it is qualifying people before they ever pick up the phone. An interactive calculator on your site that tells a visitor roughly what a project will cost, or whether they are even the right fit, does a lot of the same filtering the phone agent did for me, without anyone ever wondering if they are talking to a machine.
What I am doing now, one month on
I kept the AI agent running for the first filter, but I moved my escalation rules to be far more sensitive, any hint of frustration, complaint, or a caller who sounds over 60 (my data shows older callers hang up on AI intros at almost triple the rate of younger ones) now routes to a human within two rings instead of after a trigger word. I also softened the disclosure line so it sounds less like a legal notice and more like a normal sentence. Small changes, better outcome. The tool is not the point. Where you draw the line between machine and human is the entire project.
Frequently asked questions
Is it legal to have an AI answer business calls in the UK without telling people?
You are required under PECR rules to disclose call recording, and while there is no single blanket law forcing disclosure of AI specifically yet, treating callers honestly protects you far more than any legal minimum does, since the reputational cost of someone finding out later is higher than the cost of a short disclosure line up front.
How much does an AI phone agent cost for a small business?
Expect £100 to £400 a month depending on call volume and platform, against a typical UK receptionist cost of £28,000 to £31,000 a year including National Insurance and pension. Most small businesses will handle under 300 calls a month, which sits at the low end of most per-minute pricing.
Will an AI phone agent lose me customers?
Some, yes. In my own 30 day test, roughly 1 in 36 callers hung up once told they were speaking to an AI. That number drops if your escalation rules are fast and your disclosure line sounds human rather than legal, and it matters more in relationship-led, high-ticket businesses than in simple, information-based ones.
Should I use AI to answer calls or just add a chatbot to my website instead?
They solve different problems. A phone agent catches people who already picked up the phone, often the more urgent or higher-intent callers. A website chatbot or an interactive tool catches people earlier, before they ever call, and tends to work better for filtering price-sensitive or early-stage enquiries without any of the "am I talking to a robot" moment at all.
Related reading: AI Chatbots Are Answering Your Customers Before They Reach Your Website and Why Your Analytics Are Hiding How Many Customers AI Is Sending You.
Related: ai phone receptionist small business.
For Liverpool businesses, see AI consultancy for Liverpool businesses.
