Can an AI receptionist actually book appointments?
Three different things get sold as booking, and the difference between them is which one leaves the customer with an appointment. The question that separates them, and why a phone booking needs a different dialogue from a text one.
Sohail Safdar · · 7 min read
Yes, and "books appointments" covers three different products. The gap between them is where practices get disappointed, so it is worth being able to tell which one you are being shown.
The first distinction: a request or an appointment
Collecting a request. The assistant has a conversation, gathers a name, a service and a rough preference, and leaves your team a note: wants a cleaning, prefers Tuesday mornings. Somebody then opens the calendar, finds a slot and rings back.
That is lead capture with extra steps. It is genuinely useful, and the customer does not have an appointment. Worse, they often think they do, which you find out when they arrive.
Writing to the calendar. The assistant computes what is genuinely available, offers real slots, takes one, and the appointment exists. The confirmation is a confirmation because there is something to confirm.
Only the second removes work from the front desk rather than moving it.
The second distinction: which channel
This one is almost never made explicitly and it changes what the product is.
On a phone call, the assistant has to hold the whole booking inside the conversation, because there is nowhere else for the caller to go. There is no page to click. If it cannot offer and take a slot, the call ends with a message.
In a text or a web chat, the assistant has a third option that is often better than booking: answer the question and hand over a booking page. The customer sees the whole week at once, picks a time, and the appointment exists without anybody describing six slots in a sentence.
So "can it book?" has different right answers per channel, and a vendor whose answer does not distinguish them has probably not built one of the two.
Why a phone booking needs a different dialogue
A caller cannot see a calendar. So "when would you like to come in?" is an unanswerable question on a phone. The caller says "um, next week some time", and now the assistant has to either guess or interrogate.
The dialogue that works proposes, and the caller accepts:
"I have Tuesday at 10:20 or Wednesday at 2:40. Would either of those work?"
Two concrete options, from real availability. A volunteered preference ("anything Friday?", "mornings only") narrows the next proposal. Anything that is not a yes is treated as "not that one" rather than as agreement, because booking somebody who said "hmm" is the worse mistake, and after a few refusals it should hand the call to a person.
A text conversation needs none of that, because the booking page does the proposing visually. Which is exactly why one is not a port of the other.
Why computing availability is hard
"Find a gap in the calendar" is not it. For one appointment, the system has to know:
- Which staff can perform this service, because not everybody can.
- That person's working hours at this location, which are not the location's opening hours.
- Their existing appointments, including the one booked thirty seconds ago.
- Time off and location holidays.
- Buffers before and after: a room to turn over, notes to write.
- The duration of this specific service, not a default slot length.
- All of it in the location's own timezone, so a clock change does not move things.
Miss any one and the system offers a slot that is not free. You find out when two people arrive for the same appointment. Staff rotas and availability covers all seven.
The question that separates the products
Ask: "What happens if two people accept the same slot in the same moment?"
The three answers you will hear:
"That will not happen." It will. Two people taking the last Tuesday 2pm within the same second is ordinary on a public booking page, and it happens most when you are busiest.
"We check availability before booking." The common answer, and not sufficient. Between the check and the write, the other request runs the same check and gets the same answer. Both see a free slot. Both write. Each was correct when it looked.
"The database refuses the second one." Correct. The guarantee has to live where the write happens rather than in the code that decides to write. That is how bookings work here: availability is computed from rotas, buffers and time off, and the constraint that refuses the second write is in the database itself.
This is not pedantry. Application-level checking usually works, which is worse than never working, because it fails under exactly the load that means business is good.
What it needs to know beyond availability
A customer asking for "a cleaning next week" needs four things the calendar does not hold:
- Which service that maps to in your list, including the ones with names only you use.
- Whether they are new, since new-customer appointments are often longer and carry different paperwork.
- How long it takes and what it costs, if they ask, which they will.
- What to do when nothing fits. Offering the nearest alternative or a waitlist place beats "nothing is free, call us", which hands the work back to everybody at once.
Six things to test in a demo
- Book something. Then open the staff calendar and confirm the slot is actually blocked. This single test settles the first distinction above.
- Ask for a time that is not available. It should say so and offer alternatives, not invent one.
- Ask for a service only one person does, on a day that person is off. It should not offer it.
- Book two things back to back and check the buffer was respected.
- Cancel, then try to rebook the freed slot immediately. It should be available again at once.
- Ask it to book something it has never heard of. It should say it does not know, rather than mapping "a scale and polish" to the nearest-sounding thing in your list.
Number six is the one people skip and the most revealing, because a system that maps confidently to the nearest service will book somebody in for the wrong treatment and both sides will think it is right.
The part most systems skip: what happens after
A booking is the start of something, not the end.
- A confirmation the customer can act on.
- Reminders at useful times, to people who agreed to be texted.
- A reschedule path as easy as confirming, because the customer who cannot make Tuesday is not coming on Tuesday either way.
- A record of what changed: who moved it, when, and what it was before.
Reducing no-shows with reminders covers the second and third.
What Frontiva does, by channel
Availability is computed across staff rotas, existing appointments, time off, location holidays and per-service buffers, resolved in each location's own timezone.
Double booking is refused by a Postgres exclusion constraint: the second overlapping write is rejected by the database rather than by code that checked first and hoped. Tested with fifty simultaneous requests for the identical slot, where exactly one succeeds.
On a call, the assistant proposes real slots from that engine, takes one, and can also move or cancel an existing appointment, finding it from the number that rang rather than from anything the caller says, so there is no reference to guess at and one caller cannot reach another's booking. A cancellation asks twice, and anything that is not a clear yes keeps the appointment. That needs a voice-capable number on your own carrier account and is off until you switch it on for a location.
In a text or web chat, it deliberately has no calendar tool. It answers from your knowledge and hands over your public booking page, where the customer sees the whole week and books themselves through the same engine and the same constraint as your staff calendar. There is no second implementation for the public page, because two implementations is how they start giving different answers, and the one customers see would be the unverified one.
Two honest gaps: two-way Google and Microsoft calendar sync is not built, so a staff member's personal commitments are invisible to the rota, and charging a deposit is not connected, though the policy can be stated and shown on the booking page before the customer confirms, which is where it has to be if you want to avoid a dispute later.