AI quoting prices: the discipline before your assistant names a number

A price is a commitment the company makes, and an assistant that generates a number has made a commitment nobody authorized.

The question arrives the week the assistant starts working: a customer asks what it costs, the assistant is standing right there, and nobody has said whether it is allowed to answer. Some owners decide by enthusiasm that the machine should handle everything. Others switch the topic off entirely. Both are positions taken in a moment. Neither is a policy, and the difference shows up the first time the assistant answers on its own judgment.

Here is the framing that matters. A price is a commitment the company makes, written into its margins and its delivery capacity. An assistant that generates a number has made a commitment nobody authorized, and the customer on the other end has no way to know the difference between a stated price and an improvised one. The machine's job is to state what the company decided. It is never to decide.

So the discipline starts above the assistant, with a written answer to one question: what may be quoted? A published price may be stated. A range may be given. Or nothing may be said, and the request routes to a person. Any of those is a workable policy. What breaks is leaving the boundary to the model's imagination at the moment a customer pushes.

The failures look familiar from any sales floor, only faster. The assistant that invents a concession to close the conversation, because closing is what its training rewards. The two customers who asked the same question on the same day and received two different numbers. The price quoted from a cached page the company changed last month, honored months later because it was said. Each one is a commitment made by something that does not carry the margin.

The practice has two layers. The first is a single source of truth for everything that may be said about price, kept current, which the assistant reads from rather than remembers. The second is explicit rules for the cases between the lines: requests that exceed the stated scope, urgent timelines, volume. Each rule says either the exact posture to take or the words that hand the thread to a human.

Two quieter habits finish the job. Version awareness: a quote carries the price valid today, not the one sitting in an old page, and when the list changes the assistant's answers change with it. And a log: every figure the assistant states gets recorded, so the company can see what was promised, to whom, and when, instead of discovering its own commitments from an angry email.

We build assistants for our own funnel first, and ours do not negotiate: the dossier states the line items, the assistant explains them, and anything outside the list goes to a person. Guidance, not guarantees: what your company may quote is your pricing policy, not a technical setting, and the specifics of your obligations to customers belong to you and your advisors. The technology can enforce the boundary. Only you can draw it.