"AI receptionist" has been stretched to cover everything from a genuine voice system that books jobs to a chat bubble with a canned greeting. The term now tells you almost nothing about what you would actually be buying, which suits the people selling the weaker version.
So here is the unglamorous breakdown: what the system does, what it does not do, and the questions that separate one from the other.
What it does
Answers, immediately, every time
The first and largest function is simply that nothing rings out. No hold music, no voicemail, no "our office hours are." A caller at 11pm on a Sunday gets the same response as one at 10am on a Tuesday.
Establishes what the caller needs
Not a menu tree. A conversation that extracts the things you would have asked anyway: what has happened, where, how bad, and how soon. The output is a structured record — name, address, problem, urgency — rather than a recording someone has to listen to and retype.
Qualifies and prioritises
A burst pipe and a request for a quote on a bathroom refit are not the same call, and should not land in the same place. A working system scores urgency and routes accordingly: the emergency reaches a human now, the quote request goes into the pipeline.
Books into a real calendar
This is the part that separates a system from a message-taker. It should be able to offer genuine availability, hold a slot, and send the confirmation — without a person in the middle re-checking whether that time was free.
Follows up without being asked
The reminder before the visit, the nudge on the quote that went quiet, the review request after the job. Individually trivial, collectively the difference between a pipeline and a list.
What it does not do
This is the part most vendors leave out, so it is worth being direct.
- It does not replace your judgement on the job. It captures and routes. It does not diagnose the fault, price the work, or decide what a customer actually needs.
- It does not handle every conversation. Distressed callers, unusual situations, and anything commercially sensitive should reach a person. A system that refuses to hand off is a worse system, not a more advanced one.
- It does not fix an empty pipeline. If nobody is calling, answering faster changes nothing. That is a visibility problem and it needs different work.
- It does not run itself forever. It needs the calendar to be accurate, the service list current, and someone reviewing what it handled badly. Set-and-forget is a sales promise, not an operating model.
A system that never hands off to a human is not more advanced. It is less careful.
Five questions worth asking any vendor
- Does it write into my actual calendar, or send someone a message? These are wildly different products at similar prices.
- What happens when it does not understand? There should be a clear, fast escalation path — not a loop.
- Can I see everything it said? Full transcripts, not summaries. It is speaking to your customers in your name.
- Who owns the data? The record of your customers and jobs should be yours, and exportable, on the day you decide to leave.
- What does it do out of hours versus in hours? The honest answer is often "the same thing," which is fine — but the behaviour should be a decision you made, not a default you discovered.
How to judge whether it worked
Not by how impressive the demo sounded. By three numbers you can check yourself: how many calls went unanswered, how long it took to respond to a new enquiry, and how many enquiries turned into booked jobs. If those have not moved after a month, the system is decoration.
Which is the real test of any automation. It should be legible in the numbers you already care about, or it should not be there.