When a business decides to bring artificial intelligence into its quoting process, the first question is not "how clever is the model" but "who is doing the arithmetic". It is a technical distinction with direct commercial consequences.
The trouble language models have with numbers
Generative AI models are remarkably good at understanding natural language, context and nuance — but they are not calculators. Even the most advanced can round, confuse one formula with a similar one, or produce a plausible-looking result that is numerically wrong. On a quote, even a small arithmetic error can be the difference between a profitable job and a loss-making one.
The architectural answer: separate the responsibilities
A robust quoting system hands the calculation — waste, hours, hourly cost, margin, discount — to a deterministic engine: code that always applies the same exact formula, verifiable and testable. Artificial intelligence comes in on top of that already-correct calculation, never in place of it: it interprets a customer's request, proposes the right bill of materials, recalls the commercial history, suggests a price — but the final number always comes from the calculation engine, never from the language model's inference.
What an AI coach does well in this arrangement
With that separation, the AI can concentrate on what it is genuinely good at: reading an attached photo or drawing and pulling out the dimensions, holding a natural-language conversation to build a quote, remembering how a customer has behaved in the past, flagging patterns a person struggles to notice (margins slipping in one category, prices untouched for months). None of that goes anywhere near the arithmetic of the quote.
How to recognise a reliable system
A good test, when assessing a quoting tool with AI built in, is to ask outright: "does the language model work out the price, or a separate engine?". If the answer is not clear, the calculation probably passes through the AI as well — a risk worth avoiding in a trade with margins as tight as metal fabrication.