Using AI to Read Seal and Gasket Drawings and RFQs for Quoting

What Arrives in a Seal or Gasket RFQ

Real RFQs are messy. A buyer sends an email with a PDF drawing, a spreadsheet of O-ring sizes, a photo of a worn gasket, or a part number from a competitor's catalog and a request to match it. Some include a spec sheet with an AMS or ASTM callout, others only a trade name.

Quantities may be given as annual usage with release schedules. An estimator spends the first part of every quote simply working out what is being asked. That interpretation step is where AI helps most, because it is reading and structuring, not pricing.

• Email text with line items and quantities
• PDF or scanned drawings with notes and tolerances
• Spreadsheets of sizes and compounds
• Photos or competitor part numbers needing cross-reference

Step One: Extract Into Structured Fields

A document-reading model pulls the elements that drive cost: material and compound family, durometer, governing specification, dimensions such as ID, cross-section, OD or thickness, tolerance class, special requirements like cure date marking or FAI, and quantity breaks. Each extracted value should link to its source location on the page so the estimator can verify at a glance. Treat extraction accuracy as something to measure on your own documents, not assume.

Handwritten notes, low-resolution scans and unusual drawing formats reduce accuracy, and the system should flag low-confidence fields rather than guess.

Step Two: Match Against Your History

With structured fields, the system searches your ERP and quote archive for similar parts: same material family, similar size, same process. It returns actual yields, cycle times, scrap, tooling used and margin from those jobs. This is the part that turns tribal knowledge into something available to every estimator.

The quality of the answer depends on the quality of your history; if past job actuals are incomplete, the comparisons will be weak, and it is better to say so than to guess. For catalog items, matching is simpler: look up the price book and flag any special requirements that exceed standard terms.

Step Three: Draft the Quote, Then a Human Decides

The draft contains material cost, press or machine time, tooling, secondary operations, paperwork and margin, with each line tied to its source. The estimator adjusts for factors the system cannot see, such as customer strategy, current shop load or a risky tolerance, then approves. Nothing is sent or written to the ERP without that approval.

Well-run deployments also record the estimator's edits, which show where the model is consistently off and can improve prompts, standards or retrieval over time.

• Every line item cites its source job or standard
• Estimator edits are logged for review
• No write-back to ERP or email without approval

Privacy: Customer Drawings and Controlled Data

Customer drawings are often proprietary, and defense or aerospace drawings may be export-controlled. Sending them to a public AI service can breach NDAs or regulations. The alternative is to run open-weight models on your own hardware or a private cloud, with document permissions that mirror your existing access rules and logs of what was processed.

Netray builds this kind of private deployment on top of ERP systems such as Infor SyteLine and NetSuite, but any approach should be judged on where data goes, who can access it and how it is audited, and your compliance owner should sign off.

When This Is Not Worth It

If you receive a few RFQs a week, most are repeat parts, or you already use a quoting add-on that handles your volume, AI may add cost without much return. It makes the most sense when RFQ volume is high, estimator time is the bottleneck, drawings vary widely, and your ERP has years of usable job history. Start with a small pilot on one product family and measure time per quote, estimator edits and win rate before expanding.