AI receptionists: what they're good at, where they fall short and how to do the sums

AI receptionists are having a moment. Specialist suppliers are everywhere, and even BT Business has launched one aimed at UK small businesses.

So are they a brilliant way to catch every enquiry, or a fast track to annoying your customers? It depends on what you ask them to do. Having built one, we have some opinions.

What an AI receptionist actually is

Forget the old "press 1 for sales" menus. A modern AI voice agent holds a spoken conversation. It answers the phone, understands what the caller wants, responds using information it's been trained on and takes an action, such as booking an appointment or sending details to your team.

The Beehive Legal AI Voice Agent, which we built at Just Creative People, is a good example of the scope. It answers the firm's out-of-hours calls. It explains the firm's services, books calls through calendar integrations, follows up by email and sends leads to the CRM.

Notice that word: out-of-hours. It's covering the phone when nobody's there, not replacing anyone at the front desk.

What they're good at

• Being available. Evening, weekend, bank holiday or when your whole team is already on the phone.
• Answering the same questions. Opening hours, locations, what you do and how it works.
• Taking details consistently. Name, number and reason for calling, every time, in the same format.
• Booking. When connected to your calendar, it can offer real slots.
• Passing it on. A summary lands in your inbox or CRM, so your team starts the morning knowing who to call.

Where they fall short

Emotional or complex calls. Someone upset, confused or with a messy problem usually needs a person. A good agent recognises this and hands over or takes a message, instead of struggling on.

Giving advice. It can explain what you do, but it shouldn't tell someone what their legal position is or what's wrong with their tooth. That line needs drawing very clearly.

Getting it wrong confidently. AI can state something false in a perfectly pleasant voice. Limits on what it's allowed to say and thorough testing reduce this risk, but you can't assume it's gone.

People who just don't want it. In IONOS and YouGov's survey of UK decision-makers, loss of human touch was the top AI concern, named by 56%. Take that seriously. Some customers will prefer to leave a message or call back, and they should be able to.

Is it worth it? Do your own sum

You'll find plenty of dramatic statistics about the cost of missed calls. We've read a lot of them. They vary enormously, most come from companies selling call answering, and we wouldn't base a decision on any of them.

Your own phone log is a much better source.

Step 1: Count. Over four typical weeks, how many calls were missed? How many were outside opening hours?

Step 2: Remove the noise. Some calls are existing customers, sales pitches or wrong numbers. Estimate what share are new enquiries.

Step 3: Be realistic about callbacks. Some people leave a message or try again. What share of those would you actually have lost?

Step 4: Do the sum.

Missed new-enquiry calls a month × share you'd otherwise lose × your usual win rate × average profit from a customer = monthly profit at stake.

Here's an example with made-up numbers: 20 missed new-enquiry calls × 50% otherwise lost × 25% win rate × £300 profit = £750 a month.

Compare that with a realistic running cost, and remember the time it takes to set up and maintain. If your own number comes out at £40 a month, you may just need a better voicemail message. That's a useful result as well.

Questions to ask before you buy

1. What can it do beyond answering? Can it book, send emails and update your CRM?
2. How is it trained? On your real information, or a generic script?
3. How does it hand over? What happens with an urgent or upset caller?
4. What happens to the recordings and transcripts? Where is that data stored, for how long and who's the supplier?
5. How is pricing calculated? Per minute, per call or flat rate? What happens in a busy month?
6. Who updates it when your prices or services change?

Getting it right

Tell callers it's AI. We recommend doing this upfront. Trying to pass as human is a good way to lose trust. The ICO also expects you to be open about how personal data is used in AI systems, including why it's used, how long it's kept and who it's shared with. That guidance is under review after the Data (Use and Access) Act, so check for the current version.

Check your sector's rules. Lawyers, healthcare providers and financial firms may have additional obligations. Get these checked before you go live.

Get the voice right. Beehive Legal's callers may be worried, so the agent needed a tone that suits a sensitive subject. A chirpy voice that works for a pizza shop would be wrong there.

Start small. Out-of-hours is a great first step, since any answer beats voicemail and your team isn't displaced. Listen to real calls, fix what you hear and then decide whether to extend it.

Keep listening. Review calls regularly. They'll also tell you what customers keep asking, and that's useful for your website, too.

The bottom line

An AI receptionist won't replace a brilliant human on your front desk. It can make sure your phone is answered when that person's gone home. If your call log says that's worth something, it's worth exploring. If it doesn't, you've saved yourself a subscription.