Pricing a job with AI without hallucinating a number

A common request: "Can the AI just quote the job?" The honest answer is yes for the conversation, no for the number. Letting a model invent prices is how you end up emailing a customer a figure you can't honor.

Here's the pattern we shipped for AZ Acoustics, and reuse across the Business Logistics stack.

Split the job in two

1. The model collects scope. It runs a friendly, natural conversation — what's the room, what surfaces, what's the timeline — and calls a single structured tool when it has enough. That tool has no price fields.
2. A deterministic engine prices it. Plain code maps the collected scope to pricing bands. Same inputs, same output, every time. No model math, no hallucinated ranges.

The model never sees a price field it could fill in, so it can't get one wrong.

Defer the exact figure

Even with a deterministic engine, the right move is often to show the structure of a quote — good / better / best, what each tier includes — and confirm the exact number on a short scoping call. It qualifies the lead, protects margin, and keeps every figure honestly labelled "illustrative."

Why this beats "just ask the model"

You get the warmth of a conversation and the rigor of a spreadsheet. The customer feels heard; the business keeps control of its pricing. That's the whole game.