Language models are strong at understanding what a customer means. They are not made to calculate a long bill of materials flawlessly. One wrong quantity on a quote costs you margin or trust.
Divide the work
The best approach separates two tasks:
- The AI listens: which situation applies, which measurements does the customer mention, how many panels, rows or square metres?
- A fixed formula calculates: from that input the material follows, according to rules you checked yourself.
Example
A customer wants 12 panels on a tiled roof, portrait. The AI takes "12 panels, tiled roof, portrait" from the conversation. A calculation rule then determines how many hooks, rails and clamps belong. With a different number, the same rule simply calculates along.
What you need for it
- A list of your products with prices.
- Per common situation a rule: which items, and how many per unit.
- A place to test those rules with example values.
In Dealnote these are called calculation rules. The AI only passes on the input; the quantities come from your own formulas.