RFQ to Quotation: A Practical First AI Workflow for Manufacturers

For many manufacturers and distributors, the request for quotation is where sales time quietly disappears. RFQs arrive by email as PDFs, spreadsheets or plain text. Someone reads each one, matches items to the product catalogue, checks pricing and stock, looks at previous quotes, prepares the document, gets it reviewed and sends it. Then the CRM is updated — if there is time.

It is repetitive, rule-driven and high-volume, which makes it one of the most practical places to start with AI.

What the AI actually does

A sensible first version keeps the scope tight:

1. Reads the RFQ. Extracts customer, items, quantities, delivery dates and special terms from the email and attachments.
2. Matches items. Maps the customer's descriptions to your product codes, flagging anything it cannot match confidently.
3. Checks your data. Pulls pricing, stock and lead times from your ERP and price lists, and looks at recent quotes to the same customer.
4. Drafts the quotation. In your template, with your terms.
5. Waits for approval. A salesperson reviews, edits if needed, and approves.
6. Sends and records. The quote goes out and the CRM is updated automatically.

Nothing reaches a customer without a person saying yes.

Where the real work is

The model is rarely the hard part. The effort goes into three places:

• Item matching. Customers describe products in their own words. Building a reliable mapping between those descriptions and your catalogue, and knowing when to ask a person, is where accuracy is won or lost.
• Pricing rules. Discounts, customer-specific prices, minimum quantities and freight terms often live partly in systems and partly in people's heads. They need to be written down.
• Integration. The AI needs read access to product, price and customer data, and a way to write the finished quote and CRM record back.

Designing the approval step

The approval screen matters more than it looks. It should show the draft quote, the source RFQ side by side, and anything the AI was unsure about, clearly highlighted. The reviewer's job becomes checking and deciding, not re-doing the work.

Over time you may decide some low-value, standard quotes can go out automatically. That should be a deliberate decision made with evidence, not a default.

Starting small

It often makes sense to begin with one product line or one customer segment, run the AI alongside the existing process for a period, compare outputs, and widen the scope once the team trusts it. The goal of the first version is not to handle every RFQ — it is to handle the routine ones well and route the rest to people quickly.

What to measure

Agree the measures before building: time from RFQ received to quote sent, the share of quotes that needed significant edits, and how often items could not be matched. Those numbers tell you whether to expand, adjust or stop.