A missed call at a salon, a garage or a plumber’s is not a missed call. It is somebody ringing the next name on the list. For a business where the phone is how work arrives, the ones that go unanswered after five o’clock are the most expensive thing it does not measure.
Software that answers, holds a conversation and takes the details is now good enough to be worth considering. It is also being sold enthusiastically to people who do not need it, so here is the honest read.
General information, not legal advice. The transparency and data protection points below are the ones most often skipped by vendors. Take advice on your own situation before going live.
What it actually is
A phone number that reaches software rather than a person. It answers, speaks, listens, and does something with what it hears: takes a message, books a slot, answers a question it has been given the answer to, or decides this one needs a human and passes it on.
It is not the automated menu you already hate. That was a tree of options. This holds a conversation, and the difference is real enough that people who dread the first often do not mind the second.
It also is not a receptionist. It has no judgement, no memory of the customer who rang last week, and no ability to tell that someone is upset and needs a person. What separates a tool that answers from one that acts is worth understanding before buying either.
Where it genuinely earns its place
- Outside opening hours, where the honest alternative is voicemail nobody leaves a message on.
- Overflow. The second and third callers while somebody is already on the phone, who currently get a busy tone.
- The same four questions. Opening hours, where you are, do you do X, how much roughly. High volume, low value, and they interrupt work that pays.
- Taking details reliably. Name, number, what they want, when. A machine does not mishear a mobile number because the workshop is loud.
What those share is that the alternative is nothing. Not a person doing it worse: nobody doing it at all. That is where this pays for itself, and it is a narrower case than the marketing suggests.
Where it does not
Replacing someone who is good at the phone. If a person currently answers and customers like them, this is a downgrade being sold as a saving.
Anything with real consequences. Medical questions, money advice, complaints, anything where being confidently wrong causes harm. These systems are fluent, which is not the same as correct, and fluency is exactly what makes a wrong answer convincing.
Low call volume. If you miss two calls a week, the honest fix is a diverted mobile or a person for a few hours, not a system to administer.
The part vendors skip
Two obligations, neither of which usually appears in a sales conversation.
The caller has to be able to tell it is a machine. Under Article 50 of the EU AI Act, systems designed to interact directly with people must make that clear, unless it is obvious anyway. The design duty sits with whoever built the system, so if you buy one it is largely the vendor’s problem. Yours is the practical half: do not configure it to pass itself off as a named human colleague. Giving it a first name and letting people believe they spoke to staff is the thing to avoid, and it is also the thing that damages trust when someone works it out.
A recorded call is personal data. Voice recordings and transcripts are covered by data protection law in the ordinary way. Callers should be told, you should know where the recordings are kept and for how long, and someone should have decided that rather than accepted a default. This is separate from the AI Act and it applies whether or not any AI is involved.
Neither is difficult. Both are considerably easier to handle before launch than after a complaint.
What it costs, in shape
Pricing in this market is usually per minute of conversation, sometimes with a monthly fee on top, and it changes fast enough that any figure printed here would be wrong by the time you read it.
What is durable is the shape. Cost scales with how much people talk, not with how many staff you have, so a business with heavy call volume pays more than a quiet one for the same setup. Get an estimate based on your actual minutes rather than a headline rate, and ask what happens on a busy month.
Setup is a separate cost, and it is mostly not technical. It is deciding what the thing should say, what it must never say, and when it should give up and take a message.
Three situations, worked through
A hairdresser closed on Mondays. Calls go to voicemail and most people hang up. Something that answers, takes a name and number and says when they will be rung back is straightforwardly better than silence. Low risk, clear gain.
A garage taking bookings. Better again if it can see the diary, and this is where it stops being a phone product and becomes an integration. That is a bigger job than the demo suggests, and worth pricing as one.
A clinic with anxious patients. The calls that matter most are the ones where somebody is worried. Answering those with software is a poor trade even when it works, and this is the case for a person and an out-of-hours number.
Five things worth doing
- Count the missed calls. Your phone provider can tell you. If the number is small, stop here and save the money.
- Decide what it must never attempt. Write the list before anyone configures anything. It is more important than what it should do.
- Ring it yourself, repeatedly, including with an awkward request and background noise. Demos are recorded in quiet rooms by people who know the script.
- Make the handover to a person obvious and easy to trigger. The measure of a good one is how gracefully it gives up.
- Ask the vendor about disclosure and recordings, in writing. A blank look is information about the vendor.
Common misconceptions
- “Customers will hate it.” They hate being unable to reach you more. What they dislike is being trapped by one, which is a configuration decision rather than a property of the technology.
- “It has to sound human to work.” It has to be useful. Pretending to be a person is the part that creates both the legal question and the resentment.
- “It will replace our receptionist.” It will answer the calls nobody currently answers. That is a different and more honest claim.
- “We can set it up and leave it.” It needs someone reading what it actually said to people, at least at first. Anything running unattended eventually fails quietly, and this one fails in front of customers.
The honest summary
Worth it if the phone is how work arrives, if calls are genuinely being missed, and if the job is taking details rather than giving advice. In that case it is one of the few AI purchases with an obvious payback, because the alternative is not a cheaper process, it is a lost customer.
Not worth it if somebody already answers well, if the calls carry weight, or if you would be buying it because it is impressive. Impressive is not a business case.
If you are weighing one up, say how many calls you miss and what they are usually about. That is normally enough to tell whether it would pay for itself or just be interesting.
Sources
Article 50 of Regulation (EU) 2024/1689 as amended by Regulation (EU) 2026/1744, on transparency obligations for systems interacting with people. Checked 13 August 2026.
Common questions
Do we have to tell callers they are speaking to AI?
The system has to be built so that it is clear, unless it is obvious from the circumstances. That design duty falls mainly on whoever built it, so with a bought product it is largely the vendor’s responsibility. What is squarely yours is not configuring it to impersonate a named member of staff.
Can it book directly into our diary?
Usually yes, and this is where cost and complexity rise. Taking a message is a phone product. Writing into a live calendar is an integration, with the usual questions about what happens when two people book the same slot.
What happens when it does not understand someone?
That is the question worth pressing hardest in a demo. A good one recognises it is stuck and offers a person or takes a message. A poor one keeps trying, which is the experience that makes people give up on a business entirely.
Are the calls recorded, and does that matter?
Usually recorded or transcribed, and yes it matters. Those are personal data, so callers should be told, and someone should have decided how long they are kept rather than inheriting a default. Ask the vendor where the recordings are stored before you go live.