Quoting agent

Background

The client manufactures fasteners such as bolts and nuts and receives 30 to 50 quote requests a day from about 400 customers, handled by three salespeople and two sales support staff. Sixty percent of requests arrive by email, a quarter through the website form, and the rest by fax and phone, in every imaginable format. Some only make sense with history, like “2,000 more of the spec from last time,” and some come as PDF drawings with a spec table attached.

To produce one quote, a staff member searched the customer’s order history in the ERP, cross-checked the standard price list (updated twice a year) against prices negotiated with that customer, filled in a spreadsheet template, exported it to PDF, emailed it, and registered the quote separately in the ERP. That took over twenty minutes, and at the start of the month replies slipped to the next day. Negotiated prices lived in each salesperson’s personal spreadsheet, so quotes stalled when that person was away, and two or three times a month an outdated price list was applied, costing margin or forcing a re-send.

What we built

The goal was to remove the time spent searching, copying, and cross-checking and leave only judging prices and approving. We built five things.

1. Reading requests and mapping items

The agent extracts items, specs, quantities, delivery dates, and delivery addresses from email bodies, the web form, and attachments (spreadsheets and spec tables in PDF drawings). It normalizes notations such as “M8x30”, “M8*30”, and “8mm 30” to catalog codes. When confidence is low, it does not force a match; it shows two or three candidates for the staff member to pick. Phrases like “the spec from last time” are resolved from that customer’s recent orders.

2. History and price lookup

For each request, the agent reads the customer’s last three orders, negotiated prices, the standard price list, and volume discount tiers from the ERP. The negotiated prices scattered across personal spreadsheets were migrated into the ERP at the start of the project. Every price in the draft carries its reasoning, such as “negotiated price from Mar 2026” or “standard price × 5% volume discount,” so both the owner and a backup approver can see at once why the number is what it is.

3. The draft quote

The draft uses the spreadsheet template customers already know. Delivery dates are proposed from stock and the production schedule, and if the margin drops below the threshold, a warning appears on the approval screen.

4. Approval and sending

The staff member adjusts prices and dates on the review screen and approves. Approval triggers PDF generation, an email sent under the staff member’s name, and registration in the ERP in one step. No quote leaves without approval. If the owner is away, the quote passes to a backup approver, who can approve it as is because the reasoning is attached.

5. Follow-up

If a quote gets no reply within seven days, the owner is reminded. A dashboard shows the win rate per quote and response time per customer.

How it was rolled out

Weeks	Work
1–2	Analyzed 4,800 quotes from the previous year to build the notation patterns and item mapping dictionary. Migrated negotiated prices from personal spreadsheets into the ERP
3–5	Built request reading and history and price lookup. Ran two weeks in shadow mode alongside quotes produced by staff
6–9	Connected draft, approval, sending, and ERP registration. Two salespeople used it first while the mapping dictionary and margin rules were tuned
10–12	Added delivery date proposals, follow-up reminders, and the dashboard. Rolled out to all staff
13–14	Handover. Guide for editing the mapping dictionary and price rules, plus a monitoring screen

During the two shadow weeks, item mapping agreed with staff judgment 91% of the time, rising to 97% after the dictionary was expanded. The rest are chosen from candidates by staff. After launch, the mapping dictionary and price rules are reviewed together every quarter.

How a staff member’s day changed

Before, a sales support staff member’s morning began with opening the overnight emails, deciphering specs, and digging through the ERP. Now the review screen opens with drafts lined up in the order requests arrived, each price with its reasoning beside it. Most take three minutes to read and approve, and the time saved goes to work that needs judgment, such as negotiating prices on large orders or handling new customers.

Quotes no longer stop when a salesperson is traveling; a backup approver reads the reasoning and approves. In the four months since launch, not one quote has been re-sent because of an outdated price list.

The table below compares each task before and after.