What an AI receptionist can and cannot do
The honest version of this pitch is narrower than the marketing. An AI receptionist is very good at a specific set of calls and poor at others, and the businesses that get value from one are the businesses whose missed calls fall into the first group. Here is how to tell which you are.
What does an AI receptionist actually do well?
Answering immediately, at any hour, and taking a structured message. That sounds modest until you compare it with the alternative, which for most small businesses is a voicemail nobody leaves a message on. The single largest source of value is not sophistication, it is that the phone gets answered on the first ring at eight in the evening.
It handles the repeatable calls well: opening hours, location, whether you cover a service or an area, price ranges where they are fixed, and booking or rescheduling an appointment against a live calendar. It can also qualify: capture what the caller wants, how urgent it is, and enough detail that whoever calls back already knows the situation.
And it does not get worse when three calls arrive at once, which is when a human receptionist is at their least effective and when callers are most likely to hang up.
What does it handle badly?
Anything where the caller is upset. A distressed or angry caller needs a person, and an AI that keeps calmly asking clarifying questions makes the situation worse rather than merely failing to help. This should route to a human immediately, and if nobody is available it should say so plainly rather than continue.
Anything requiring judgement about an exception. Whether to waive a fee, whether to squeeze in an emergency job, whether this particular customer's complaint warrants a refund. Those are the calls where getting it wrong is expensive, and they are exactly the calls where a confident wrong answer is worst.
Complex diagnosis, too. A caller describing an unusual technical problem is not well served by something that has to map their description onto a fixed set of intents. It can take the details accurately, which is genuinely useful, but it should not attempt an answer.
| Call type | AI receptionist | Why |
|---|---|---|
| Hours, location, coverage | Handles | Fixed answers, high volume, low risk |
| Book or reschedule | Handles | Structured task against a live calendar |
| Qualify a new enquiry | Handles | Capturing detail is the whole job |
| Quote a fixed price | Handles | Only where the price genuinely is fixed |
| Quote a variable price | Hands over | A wrong number becomes an expectation |
| Upset or distressed caller | Hands over immediately | An AI persisting makes it worse |
| Exception or goodwill decision | Hands over | Commercial judgement, expensive to get wrong |
| Unusual technical problem | Takes details only | Should not attempt a diagnosis |
Which businesses get the most from one?
Ones where a missed call is a lost job and the caller will simply ring the next name on the list. Trades, clinics, salons, repair services, lettings, professional services with an appointment model. If your enquiries arrive by phone and your competitors are three seconds away in the same search results, answering first is most of the sale.
Also businesses with a spiky call pattern. If you get twenty calls in an hour after an advert runs and four the rest of the day, staffing for the peak is uneconomic and staffing for the average loses the peak. That is the clearest case there is.
It matters least where callers are few, high-value, and known to you. If you take six calls a week and each is a long conversation with an existing client, an AI answering first adds friction rather than removing it.
How do you judge whether it paid for itself?
Count what you are missing before you buy anything. Most businesses do not know their missed-call rate, and it is the number the entire decision rests on. Your phone provider can usually report unanswered and out-of-hours calls; get that first.
Then instrument three things afterwards: calls answered that would previously have gone unanswered, of those how many produced a booking or a qualified enquiry, and how many were handed to a human. That last one is not a failure metric, it is a design metric. A handover rate that is very low means it is probably overreaching.
Compare the result against the value of one job, not against the subscription. For most businesses with a meaningful average job value, the arithmetic is decided by a small number of recovered calls, which is why the missed-call baseline matters more than any feature comparison.
What should callers be told?
That they are speaking to an automated assistant, at the start. Some vendors treat disclosure as a conversion problem to be minimised. It is not: callers work it out, and the ones who feel misled are the ones who complain about it publicly.
There is a regulatory dimension too, and it is moving. Several jurisdictions now require disclosure when a person is interacting with an AI system, and rules around recorded and synthesised voice calls vary by state in the US as well as by country. If calls are recorded, that needs its own consent, and two-party consent states have stricter requirements than others.
Practically: disclose up front, offer a route to a person in the first few seconds, and never imply the caller is speaking to a named member of staff.
Common questions
- Will callers know they are talking to an AI?
- They should be told at the start, and most work it out regardless. Disclosure is also increasingly a legal requirement rather than a courtesy, with several jurisdictions now mandating it when someone interacts with an AI system, and recorded-call consent rules varying by state and country. Callers who feel misled are the ones who complain publicly, so there is no upside in hiding it.
- Can an AI receptionist book appointments?
- Yes, and it is one of the tasks it does best, because booking is a structured job against a live calendar rather than an open-ended conversation. It can also reschedule and cancel. What it should not do is make exceptions to your booking rules, such as squeezing in an emergency, since that is a commercial judgement.
- What happens when the AI cannot help?
- It should hand over to a person and say so plainly, and if nobody is available it should say that rather than continuing to ask questions. A handover is a designed outcome, not a failure: a very low handover rate usually means the system is overreaching on calls it should be passing on.
- How do we know whether an AI receptionist is worth it?
- Start by measuring your missed-call rate, including out of hours, which your phone provider can usually report. That baseline is what the decision rests on. Afterwards, track calls answered that previously would not have been, how many became bookings or qualified enquiries, and the handover rate. Compare against the value of one job rather than against the subscription cost.
- Is an AI receptionist suitable for a small business?
- Often more suitable than for a large one, because the alternative is usually voicemail rather than a staffed switchboard. It fits best where a missed call means the caller rings a competitor, and where call volume is spiky enough that staffing for the peak is uneconomic. It fits worst where calls are few, long, and from clients who already know you.
