Four thousand calls a month

Earlier this week I demonstrated an AI receptionist to a staffing agency that takes roughly four thousand phone calls a month. That is about a hundred and eighty a day — one every two and a half minutes, all day, every working day. Someone is answering that phone instead of doing anything else.

The interesting part of building it was not the voice. It was the very first decision the receptionist has to make, before it can say anything useful at all.

The first question is who

The same number is called by two completely different people. A client has a shift to fill. A candidate wants work. They want opposite things, they use different vocabulary, and the wrong opening question wastes everyone’s time.

So the receptionist works out who it is talking to first — and then runs a different conversation depending on the answer. That sounds obvious written down. It is the thing that makes the difference between a phone tree and something worth calling.

A client wants a mandate opened

When it is a client, the receptionist helps them create a new mandate: what is needed, where, when, for how long. The conversation ends with a request that exists in the system, not a note that someone has to key in later. The call is the entry.

A candidate wants a shift

When it is a candidate, the receptionist helps with booking and matching — what is available, what fits, what they can take. Again the outcome is a record, not a callback promise.

This is the same distinction I keep coming back to: an AI that converses is a novelty. An AI that converses and leaves a transaction behind is a colleague.

It had to sound like here

The part the agency reacted to most in the demo was not the logic. It was the French.

We got the receptionist speaking a genuinely Québécois French — the accent and the turns of phrase people here actually use. A caller in Trois-Rivières should not feel like they have been transferred overseas. It is a small thing that stops being small the moment you hear it done badly.

A receptionist who sounds like she is from somewhere else is a receptionist people stop trusting on the second sentence.

What we are still building

Two things are in progress and worth being straight about, because neither is finished.

• Bilingualism. She is good in French. Making her switch cleanly between French and English mid-call, the way a bilingual human does without thinking, is the work under way now.
• Escalation. Some calls should not be handled by an agent at all, and the receptionist needs to recognise those and hand them to a person — warm, with the context already gathered, not a cold restart. That is coming.

What it means for the agency

Four thousand calls a month is not a communications problem. It is a capacity problem. Every one of those calls currently interrupts somebody who was doing something else, and most of them are routine: a shift to fill, an availability to confirm, a question with a known answer.

Taking the routine ones off the humans gives the agency back a very large amount of time — and the calls that genuinely need a person get someone who is not already on their fortieth call of the day.

This is heading to deployment shortly. If your team spends its day on the phone, that is the part worth looking at first. Book a demo and we will show you the receptionist handling your kind of calls.