Reading RFQs and Prints
Every fabricator has a morning routine of opening emails, downloading attachments, and typing part numbers, quantities, materials and notes into a quote form. An AI agent can do the first pass: pull the customer, requested quantities and due date from the email, read the title block and notes on the print for material, finish and tolerance callouts, and flag what is missing. It can also check whether the part number or a similar one exists in your ERP.
The estimator still prices the job, but starts from a structured summary rather than a blank form. Accuracy on messy or hand-marked prints varies, so every extraction should show its source for quick verification.
Checking a Quote Against History
A second set of eyes on every quote is valuable and rarely affordable. An AI layer connected to your ERP can find comparable past jobs by material, thickness, size and process, and compare their actual cost and margin with the draft.
- Flag drafts well below the actual cost of similar jobs
- Show past quantities and prices for the same customer and part
- Surface outside-processing costs and vendor lead times from earlier jobs
- Note engineering changes or rework that hit earlier orders
Asking the ERP Instead of Building Reports
Owners and operations leaders often wait on a report. Natural-language access to the ERP answers questions directly: which open jobs are past due, what is on hand for a given material, which customers have the lowest margin, how many hours the brake ran last month. The assistant should run read-only queries, show what it ran, and respect existing user permissions.
This works with JobBOSS, E2, Epicor, Global Shop, SyteLine, NetSuite and others, subject to how the data can be accessed. Answers are only as good as the underlying data, and an assistant will not fix mislogged hours.
Try it here, free
Sheet Metal Quote Review Checklist Before Sending
A grouped checklist that catches the common quoting misses on drawings, material, cutting, forming, hardware, finishing and terms.
Keeping Customer Data Private
204-7012 and CMMC. Sending those drawings to a public AI service can breach contracts or regulations. A private deployment runs open-weight models on hardware in your facility or a controlled private cloud, with retrieval over your own files and logs of what was asked.
That architecture helps you meet your obligations, but compliance remains your responsibility, and you should confirm requirements with counsel or your contract officer.
When AI Is Not Worth It
If you quote a handful of jobs a week, run mostly repeat orders, or your estimator is not a bottleneck, AI adds little. If your ERP data is thin, because jobs are not closed out or time is not captured, fix that first. And if the pain is on the floor, such as machine downtime or tooling changeovers, the answer is more likely a scheduling change, a new press brake tool library or maintenance, not a language model.
Pick a use case with measurable results, run it for a month, and compare quote turnaround and accuracy before expanding.
Frequently asked questions
How can a small fabrication shop use AI?
Start with RFQ intake and quote checking, where time is lost and results are measurable. Another practical use is natural-language questions over your ERP so you stop waiting on reports. Avoid broad projects before you have a clear task.
Is it safe to upload customer drawings to ChatGPT?
Often not. Customer NDAs, export-controlled drawings under ITAR or EAR, and CUI obligations can prohibit it. Check your contracts. A private deployment inside your own network avoids sending drawings to a third party.
Can AI read a DXF or PDF print and price it?
It can extract notes, materials and dimensions from many prints and help draft a quote, but pricing should come from your calculators and history with an estimator approving. Treat the AI as an assistant, not an autonomous quoting engine.
Netray builds private AI for fabrication shops on top of your ERP and drawings, starting with one use case such as RFQ intake or quote checking, with human approval throughout.