AI receptionist for your business: what it can and can't do

What an AI receptionist actually does on a call

Someone rings your number. Instead of voicemail or a hold queue, they get a voice agent that's been set up with your services, your opening hours, the prices you're happy to share and the questions you use to qualify a new enquiry. It can:

• Answer common questions from the information you give it, so what you do, where you are, when you're open and how the process works.
• Qualify the caller with the same questions your team asks, and write down the answers.
• Book the appointment into your calendar while they're still on the line, and send a confirmation by text or email.
• Put the call through to the right person when the caller asks for someone, or when the conversation needs a human.
• Log everything. The recording, a written summary and the answers go into your CRM on the contact, so nobody has to type up notes.

The agent answers calls, and it doesn't make them. Appointment reminders and follow-up to enquiries that didn't book go out by text and email through our follow-up and nurture work, and never by AI calls.

Where it works well, and where it needs a person

Call type	Good fit for a voice agent?	Why
Out-of-hours and overflow calls	Yes	The other option is voicemail, and most callers don't leave one
New enquiries with standard questions	Yes	The questions are predictable and the goal is a booked call
Appointment booking and rescheduling	Yes	Clear rules, a calendar to check and a confirmation to send
Reminder and follow-up calls	No, we don't build these	Reminders and follow-up go by text and email instead
Complaints and upset callers	No	Pass them to a person straight away
Pricing negotiations or bespoke quotes	No	Take the details and book a call with your team
Anything regulated or sensitive	No	Medical, legal or financial advice needs a qualified person

Our rule of thumb: let the agent take the calls where the right answer is already written down somewhere in your business. When the answer needs judgement, the agent's job is to get the details and get the caller to the right person fast.

What can go wrong, and how we stop it

The agent says something wrong. A business is responsible for what its automated systems tell customers. In Moffatt v Air Canada (2024), a Canadian tribunal held the airline liable after its website chatbot gave a customer the wrong information about a fare policy. So what we do is keep the agent to answers from information you've signed off, and when it doesn't know, it says "I'll get someone to confirm that."

Callers feel tricked. Tell callers at the start that they're speaking to an automated assistant, and give them a way to reach a person. People are a lot more patient with a voice agent that's honest about what it is.

Automated calls break the rules. Under the UK's PECR rules, automated marketing calls need the person's specific prior consent. General marketing consent isn't enough. The voice agents we build only answer calls, and reminders and follow-up go by text and email instead.

Data ends up in the wrong place. Calls get recorded and transcribed, so the recordings and summaries need to sit in accounts you control, and only the people who need them should have access.

How we build one

1. Listen first. We go through real calls with you, so the common questions, the qualifying questions, and the calls that should always go to a person.
2. Write the call flow. What the agent says, what it asks, when it books and when it hands over. You sign it off before anything goes live.
3. Connect it. Your phone number, calendar and CRM, all built in your own accounts.
4. Test it on real situations. Normal calls, awkward calls, people who go off script, bad lines and people who ask for a human.
5. Launch in stages. Usually out-of-hours calls first, and then overflow, and then more once the summaries show it's handling calls properly.

There's more on what the service covers on our voice agents page, and our approach page goes through each stage. If most of your enquiries come in writing, a chatbot might be the better first step, so have a look at custom vs off-the-shelf chatbots.

What it costs

There are two parts:

• The build. Projects start at £4,500. The price depends on how many call types the agent handles and how many systems it connects to. You get a written scope and price before we build anything.
• Running costs. Phone minutes and AI usage get billed per minute by the providers, and you pay them from your own accounts. The total depends on how many calls you get and how long they are, and we work it out with you from your current call numbers before you commit.