How to set up AI drafted quotations with a named approver

List every field on the quotation form, then sort each into one of two piles: facts held in your item master, and promises that only a person can make.

Catalogue facts are safe to fill. Part number, drawing revision, unit of measure, standard rate and tax code all come from the item master that your accounts and stores teams already maintain. Promises are not safe.

Discount, delivery date, payment terms, minimum order quantity and any claim about fitness for a particular use stay blank until a human completes them. A lead time copied from last year looks harmless, yet it becomes a dispute once the customer holds the PDF and your stores cannot ship the casting. Blank is better than wrong.

Drafting begins with the customer’s request for quotation, which may arrive as a scanned PDF, a spreadsheet with a bill of materials or three lines typed into a mail. The software extracts items and quantities.

Each requested line is matched to the catalogue by description and part number, with the drawing revision checked where the enquiry quotes one. A weak match gets no guess. The line is marked unmatched, and the two closest candidates are offered, so the reviewer picks one in a click instead of searching the whole catalogue by hand.

Rates come from a price list lookup. Nothing is calculated or invented. An item with no rate stays empty, so the gap shows on the screen instead of hiding inside the grand total of a quotation that looks complete.

Name one approver per quote band. Small routine quotes may go to a sales coordinator, while large or custom ones go to the sales head or the owner. Write the bands down.

The approver sees the draft beside the original enquiry, with unmatched lines and blank promise fields highlighted so that nothing needing a human decision can be skimmed past. They fill the discount and the delivery date, check the freight terms and incoterm, then confirm or reject the draft with a note that the sales coordinator can read.

Never release from the drafting screen. Release needs an approval record with a name and a time, so a disputed quote traces back to the person who signed it. Give each band a deadline too, because a quote parked two days with the sales head loses to a competitor who answered the same enquiry that same morning.

Once approved, the system builds the PDF on your letterhead and sends it from the sales mailbox to the buyer who raised the enquiry. A filed copy carries the approver’s name.

Log every edit the approver makes. When the same description keeps changing or the same unit keeps getting fixed, the catalogue entry needs attention, not the software. That is a master data problem. Version the price list and tag each released quote with the list it used, so any old quotation can be reproduced exactly as it left the building.

Resist switching on the whole catalogue. Pick one product family with clean master data and a steady flow of enquiries, such as fasteners, cable glands or hydraulic fittings, and begin there.

For a fortnight, let your team keep quoting by hand while the model drafts the same enquiries in parallel, so nothing the model drafts reaches a customer. Compare each draft with the manual quotation and count the corrections by field and by customer, which shows which catalogue entries and which buyers cause the most rework.

This week, write down every quote field and mark each as fill or leave blank, using a pencil and the last ten quotations you sent. That sheet becomes your specification.