How it works
What actually happens when somebody calls.
Six steps, in the same order every time. You keep your number, you set the hours and the rules, and Callahand picks up when you cannot.
The call
From the first ring to your inbox.
The call reaches your receptionist
Your number rings as it always has. When nobody picks up, or it is outside the hours you set, the call reaches Callahand instead of a voicemail beep.
It says what it is
The receptionist identifies itself as an AI assistant before anything else. Callers are never led to believe they are speaking to a person, and the disclosure is not a setting that can be switched off.
It works out what they need
It asks what the job is, matches it against the services you offer, and asks again rather than guessing when the answer is unclear. If a caller describes a hazard, that takes over the call immediately.
It checks your real availability
Times come from the calendar you connected, filtered by the hours, buffers and appointment lengths you configured. It reads two or three aloud rather than a list nobody can hold in their head.
It books, and only says so once it is true
When the caller picks a time and confirms it, the appointment is written to your calendar. If that write fails, the caller is told plainly that somebody will follow up — not that a time is held.
You find out what happened
A written summary reaches you after the call: who rang, what they needed, how urgent it sounded, what was booked, and what still needs a person. The structured record is in your dashboard.
Before the first call
You tell it how your business works.
Setup is a guided sequence you can leave and come back to. Nothing is guessed on your behalf, and the receptionist does not go live until you publish it.
Your services and your hours
What you do, how long each job usually takes, when you work, and how much notice you need before an appointment.
Where you go
The areas you cover, so a caller outside them is told honestly rather than booked into a job you would have to cancel.
What counts as an emergency
The conditions specific to your trade, and what should happen when one comes up. This is the part we ask you to read twice.
Your calendar
Connected once, with the narrowest access the booking needs. Availability is read from it; appointments are written to it.
How you want to hear about it
The address a call summary should reach, and who else on your team gets one.
How it should sound
The greeting, the tone, and what it should say when it cannot help — reviewed by you before anything goes live.
The parts that are not negotiable
Where the product refuses to be clever.
Most of what makes an AI receptionist safe is a decision to keep the model out of the decisions.
- The model never decides anything that matters. Understanding what a caller said is a language problem. Whether to book, whether something is an emergency, whether a caller may change a setting — those are decided by ordinary code with tests, not by a model.
- A caller cannot talk their way past a rule. Somebody claiming to be the owner, the developer or an administrator gets no more authority than anybody else. There is no phrase that unlocks anything.
- Safety overrides the script. An emergency is matched against the whole conversation on every turn, and once raised it cannot be talked back down. Ordinary intake stops there.
- A failure is never dressed up as a success. If the calendar is unreachable, if a tool times out, if the receptionist is unsure — the caller is told, and you are told. Nothing is reported as done because it probably worked.
Want to see it on your own calls?
Tell us what your trade is and how your calls come in. We will walk through what Callahand would do with them, step by step, on a real call flow rather than a slide.