What a $0.000008 brain could do for a voice receptionist

A voice receptionist is four things: a phone number, a voice platform that turns speech into text and back, the call flows that decide what happens when, and the AI “brain” that reads each moment and picks the next move. Most of the mystique — and most of the fear about cost — sits in that last piece. So let’s put a number on it.

On September 16 we ran a real routing decision through a cheap inference provider — the kind of one-word call a receptionist brain makes dozens of times per conversation (“is this a booking, a question, or something else?”). Measured result: 137 tokens total, billed $0.000008, answered in three to four seconds. That’s eight millionths of a dollar, for the actual thinking.

For context, here are live market prices from the same day: the cheap model we used charges $0.046 per million input tokens and $0.092 per million output tokens. A premium-class model on the same market lists $0.80 and $4.80 — roughly fifteen to twenty times more for the same unit of thinking.

Now scale it to a real call. A typical receptionist call makes about ten of these decisions, with a few hundred tokens of conversation history attached each time. Our worked estimate: roughly $0.0003 per call in AI cost. A business taking 300 calls a month would spend on the order of ten cents a month on the brain. These are estimates, not invoices — your call lengths will differ — but the order of magnitude is the point: the AI is the cheapest line item on the call, not the scary one.

Here’s what cheap inference changes, and what it doesn’t. It changes the per-call cost — dramatically. It does not change the call flows, the human review of every delivery, the failover checks, or the monthly monitoring. A cheap brain with good guardrails beats an expensive brain with none, every time. The quality of a receptionist lives in what it’s allowed to say and what happens when it doesn’t know — not in how expensive its thoughts are.

That’s also why we’re relaxed about telling you all of this. Our prices are fixed regardless of which model wins our test script for your business. If the cheap brain passes, you keep the savings in your running costs. If it doesn’t — and sometimes it doesn’t, which is a whole future post — we’ll tell you that too.