AI receptionist: how to stop losing missed calls
An AI receptionist is a voice assistant that answers when nobody else picks up: out of hours, when the line is busy or after a set number of rings. It says it's an AI, finds out what the caller needs, books them into your calendar or logs the enquiry in your CRM, and hands over to a person when needed.
How a voice agent works inside, why response time matters and where the technology stops are covered in What is an AI voice agent?. This article is more practical. It deals with one job, the inbound calls nobody at your business picks up, and how to set up an AI receptionist so it catches them without embarrassing you.
I’m writing from practice: we’ve run 1,700,000+ AI calls over 3.5 years since 2022.
How many calls do you actually miss?
Most owners don’t know the exact number. Missed calls get buried in the call log on your mobile. Some people ring back and some don’t. Before you change anything, measure it:
- Keep a record for two weeks. Use the itemised list from your phone provider or your phone system, or the missed calls on your mobile. For each one, note whether it was a known number or a new one, and whether you called back.
- Sort them by time. Working hours when you were busy, evenings, weekends. Each group needs a slightly different fix.
- Count the new numbers you never heard from again. Those calls most likely went to a competitor.
To put a rough figure on what those calls cost you, use the missed-call calculator. It works only with the numbers you enter yourself.
What an AI receptionist does
- Answers and introduces itself. It says it’s your business’s AI assistant and asks how it can help.
- Finds out what the caller needs. Name, phone number, which service, the address or the make of car, how urgent it is. It asks one question at a time, not like a form.
- Answers common questions. Opening hours, where to park, whether you do a particular job, what to bring. It uses only the material you give it.
- Books into your calendar. It offers free slots, saves the booking and sends a confirmation by text.
- Hands over to a person. It transfers the call, or arranges a callback and sends you a summary.
- Logs the outcome. It saves a summary, the transcript, a recording if you use one, and a task for you on the contact in your CRM.
What an AI receptionist doesn’t do
- It doesn’t know what you haven’t told it. If the answer isn’t in its material, it shouldn’t guess. A well set-up receptionist says it will find out and creates a callback task.
- It doesn’t promise prices or finish dates unless you’ve given them to it. A tradesperson can’t price most jobs without seeing them anyway.
- It doesn’t resolve complaints. It takes them down, asks for the key details and passes them to a person.
- It can’t see your workshop or treatment room unless it’s connected to a system that holds that information. It can only answer “Is my car ready?” if you record job status somewhere the AI can read.
- It doesn’t replace your callback. If a caller wants to speak to the owner, the receptionist arranges it. Someone still has to ring them.
When should the AI receptionist take the call?
This is the most important decision at the start. There are several modes, and you can combine them:
| Mode | When the AI answers | Who it suits |
|---|---|---|
| Out of hours | after closing time, at weekends and on bank holidays | any business that gets calls after closing |
| Overflow | nobody answers within a set number of rings, or the line is busy | trades, garages, salons, anyone whose hands are full |
| Selected lines | only one number, such as a bookings line | businesses with more than one number |
| All day | every call, with people taking only the handovers | businesses where most calls are bookings |
I’d start with out of hours. The risk is smallest, because otherwise nobody would answer at all, and within a few weeks the transcripts show you what callers want and how the AI talks to them. Then add overflow during working hours.
Setting it up on your number
You don’t need a new number. Calls usually reach the AI receptionist through call diversion from your existing number: a conditional divert set with your phone provider (on no answer, when busy, when unreachable), or rules in a cloud phone system. With overflow, watch the number of rings. Too many and the caller hangs up before the AI answers. Too few and you don’t get the chance to pick up yourself. Test it on a few calls. Our phone system page explains how we handle this in Clientee CRM.
Opening hours and bank holidays
The receptionist has to know when you’re open, when you’re on holiday and when it’s a bank holiday. Otherwise it will offer slots when nobody is there, or tell callers you’re open when you’re not. In the UK this needs a little care, because England and Wales, Scotland and Northern Ireland have different bank holidays, and GOV.UK lists them for each. Keep your opening hours in one place that both the calendar and the AI read from, and enter holidays as blocked time in the calendar.
Handing over to a person
This is where callers decide whether they’re happy. Settle these points in advance:
- When the AI hands over. The caller asks for a person, it’s a complaint, the topic isn’t in the material, the caller is upset, or it’s urgent (a burst pipe, a car broken down on the motorway).
- To whom. A named person, or an order of people based on the rota. In a small business that’s often the owner and one stand-in.
- How. A direct transfer, or an agreed callback. A transfer only makes sense when someone is free to answer.
- What happens if nobody picks up. The AI goes back to the caller, apologises, promises a callback at a specific time and texts you a summary. A call that ends in endless ringing is worse than no AI at all.
Booking straight into the calendar
Booking during the call is what sets an AI receptionist apart from voicemail. For it to work, the calendar needs clear rules:
- How long each service takes. An oil change 30 minutes, an MOT an hour, a haircut 45 minutes. The AI only offers times where the service fits.
- Gaps between appointments for travel or cleaning up.
- Who and where. If you have several engineers, stylists or treatment rooms, the calendar has to know who does what.
- What shouldn’t be booked straight in. Large jobs deserve a site visit or a consultation first. The AI books that instead.
- Sync. If appointments also live in a Google or Outlook calendar, they must sync both ways, or the same slot can be booked twice.
After booking, the caller should get a text confirmation and a reminder the day before. Our booking calendar page shows the calendar the AI books into.
Logging the call in your CRM
A call that leaves behind only “someone rang about a repair” isn’t much use. The contact record should hold:
- name and phone number, plus email and address where relevant
- the reason for the call and how urgent it is
- a short summary (three lines, not the whole transcript)
- the transcript, and a recording if you keep them, for checking
- the outcome: booking, enquiry to quote, question answered, handed to a person
- a task for a named person with a due date, if someone needs to get back to the caller
Each new enquiry should land straight in your sales pipeline so it doesn’t get lost. Why the speed of your next reply matters is covered in How to get more leads.
What the start of a call can sound like
An illustration for a garage, not a transcript of a real call:
AI: Hello, you’re through to the AI assistant at Northgate Garage. This call is recorded. How can I help?
Caller: I need my tyres changed over.
AI: Of course. Are they on their own wheels, or do the tyres need fitting to your existing rims?
Caller: On their own wheels.
AI: The next free slots are Thursday at 8am or Friday at 2:30pm. Which suits you?
The first sentence says it’s an AI and that the call is recorded. After that, it gets on with the job. The opening shouldn’t take more than a few seconds.
What to check on the legal side
This isn’t legal advice. It’s a list of points to go through with your provider, and with a solicitor if you’re unsure.
EU AI Act: callers must know they’re talking to an AI
Article 50(1) of the EU AI Act requires providers to design AI systems that interact directly with people so that those people are informed they’re dealing with an AI system, unless that’s obvious from the circumstances. Article 50(5) says the information must be given clearly, at the latest at the first interaction. Under Article 113, these rules have applied since 2 August 2026.
The Digital Omnibus on AI, Regulation (EU) 2026/1744, came into force on 27 July 2026. It mainly pushed back deadlines for high-risk systems. It did not change the duty in Article 50(1) to tell people they’re talking to an AI.
If you’re a UK business with callers in the EU, assume the rule applies to you. If all your callers are in the UK, the AI Act isn’t what binds you, but we set every receptionist up the same way anyway. In practice the AI says it’s an AI in its first sentence. Don’t present it as “Jane from reception”.
UK GDPR: every call contains personal data
UK GDPR and the EU GDPR use the same article numbers for the points below, so the checklist works for both.
- Lawful basis. You need one under Article 6 for using the details from a call. For handling an enquiry or a booking, that’s usually contract: the ICO’s guidance says this basis covers steps a person asks you to take before entering into a contract. For recording calls to check quality, businesses often rely on legitimate interests, and the ICO’s own example is a company using helpline recordings for staff training. Even then you still have to pass the necessity and balancing tests. Choose your basis and write it down.
- Telling the caller. Under the right to be informed (Article 13), people must know who is processing their data, why, on what basis, who receives it and how long you keep it. On the call, a short line about recording is enough, with a privacy notice on your website for the rest.
- How long you keep it. The ICO says you shouldn’t keep personal data indefinitely “just in case”. Set a retention period for recordings and transcripts and switch on automatic deletion.
- Who processes the data. AI calls run through language model and telephony providers, often outside the UK and EU. Clientee CRM runs on HighLevel, which stores data in the US and is certified under the EU-US Data Privacy Framework and its UK Extension. List these processors in your privacy notice.
- Health and other sensitive data. Clinics need to expect callers to volunteer details about their health. Under UK GDPR that’s special category data, and using it needs an Article 9 condition on top of your lawful basis. We don’t put health information into the AI’s material or its answers. I’d set the script up so the AI doesn’t ask about medical conditions and passes those calls to a person.
Outbound calls
An AI receptionist deals with inbound calls. If you also want the AI to ring people, marketing calls made by an automated system fall under stricter rules in PECR. We cover them in What is an AI voice agent?.
Where AI receptionists usually go wrong
The technology is rarely the problem now. Problems come from the setup and from what happens after the call.
- A long introduction. Three sentences about the business and data protection before the AI asks what the caller wants. People hang up. Cut the opening to one sentence and put the detail in your privacy notice.
- Answers from outside the material. An AI that makes up a price or bank holiday opening hours does more harm than good. Its instructions must say clearly what it must not answer.
- Out-of-date material. You changed your prices or opening hours and the AI doesn’t know. Its material belongs in the same routine as updating your website and Google profile.
- A calendar that doesn’t match reality. An engineer shows as free, but is on an all-day job nobody entered. The AI then books a customer into a slot you can’t keep.
- Handing over to nobody. The AI transfers the call to a number nobody answers. As above, there must always be a fallback.
- Nobody reads the summaries. This is the most common failure. The AI logs the call perfectly and the enquiry then sits there for three days. Every log needs a task for a named person and a notification on their phone.
- Nobody listens to the calls. Read the transcripts every day for the first few weeks, then spot-check at least once a week. You’ll see where calls get stuck and which questions the AI has no answer for.
Test how natural the voice sounds, how it copes with interruptions and whether it gets names right. The section on accents and names in What is an AI voice agent? lists what to try.
How to get started
- Measure for two weeks. Count missed calls by time of day and by type of caller.
- Choose a mode. Start with out of hours, then add overflow.
- Write the material: 15 to 20 of your most common questions with short answers, your services with how long each takes, what the AI must not say and when it hands over.
- Connect the calendar and CRM so bookings and summaries are created automatically.
- Test it. Ring it yourself, get colleagues to ring it and be awkward on purpose. To get a feel for how it handles customers, try the AI receptionist demo: enter your website and chat with it as a customer would.
- After launch, read the transcripts and adjust the material based on what people actually ask.
How we set up AI receptionists for clients, and what gets written to the CRM along the way, is on our AI voice agents page.
Sources
- Regulation (EU) 2024/1689 (AI Act), consolidated text of 27 July 2026, Articles 50 and 113 (EUR-Lex)
- Regulation (EU) 2026/1744 amending the AI Act (Digital Omnibus on AI) (EUR-Lex)
- A guide to lawful basis (ICO, UK GDPR guidance)
- What is the 'legitimate interests' basis? (ICO)
- Right to be informed (ICO, UK GDPR guidance)
- Principle (e): Storage limitation (ICO)
- Special category data (ICO)
- HighLevel Inc: EU-U.S. Data Privacy Framework and UK Extension participation (U.S. Department of Commerce)
- UK bank holidays (GOV.UK)