Short answer: You speed up RFQ quoting in a job shop by automating the preparation work around each estimate and leaving the estimate itself to your estimator. AI can open the RFQ email, read the drawings and notes, pull out material, quantities, tolerances, finishes and due dates, and find similar parts you have quoted before. Your estimator then starts from a filled-in quote sheet and spends their time on setups, cycle times and price, which is where their judgement actually matters.
The problem
Ask anyone who quotes for a machine shop or fabrication job shop where the day goes and you hear the same story. An RFQ arrives as an email with a zip file of PDF drawings, maybe a STEP model, and a line about quantities and a needed-by date. Someone has to open every file, work out what the part is made of, spot the tight tolerances, the heat treat, the plating and the inspection requirements, check material prices, look for a similar job from last year, and only then start estimating.
Shop people describe this openly. In a long-running eMastercam forum thread on quoting processes, one programmer explains that he sometimes writes quick toolpaths just to get a cycle time for a quote, which can take an hour and eats into his programming work. Others say they avoid that by estimating from similar past parts, and focus their attention on close tolerances, special processes and any tooling they would need to buy. The common thread is that quoting falls on the same few experienced people who are also needed on the floor.
The problem has not gone away. In July 2026 the CAM software maker CloudNC described quoting as a bottleneck that depends on a small group of experienced people, who also handle production, programming and customer issues, so RFQs queue up behind everything else. Developers are still trying to solve the dull part: a Show HN post from May 2026 for a tool that reads PDF, DXF and STEP files into shop quotes noted that reconciling dimensions on the PDF drawing with features in the 3D model was the hardest thing to get right.
What it costs
Slow quoting costs you twice. First in time: the people who can estimate well are expensive. The US Bureau of Labor Statistics puts the median wage for cost estimators at $37.86 an hour (May 2025), and in a small shop the estimator is often the owner or a senior machinist instead. BLS puts the median machinist wage at $58,750 a year, and every hour they spend on paperwork is an hour away from a machine.
Second, in work you never see. Most quotes are lost. A sponsored article in Modern Machine Shop, written to promote the quoting software company Paperless Parts, says most shops win roughly a third of the jobs they quote. That is a vendor figure, so treat it as a rough guide, but most shop owners will recognise it. If two out of three quotes go nowhere, the time spent on each one matters a great deal, and a quote that arrives a week late is very likely one of the losers.
Why it keeps happening
The inputs are messy
Every customer sends RFQs differently. One sends a clean PDF drawing and a STEP file, another sends a scanned print with handwritten revisions, a third pastes a parts list into the body of an email. There is no standard format, so the first half hour of every quote is sorting and reading.
The knowledge is in people's heads
The shop's real pricing knowledge sits with one or two people: which customers pay for quality, which materials are hard to get, which features always cause trouble on a particular machine. That makes them the bottleneck by design.
Past quotes are hard to find
Old quotes live in spreadsheets, the ERP, email and filing cabinets. Finding "that bracket we did for the pump company in 2024" depends on someone remembering it. The MMS piece above makes the same point: past estimates stored across drives and cabinets are rarely reused because nobody can search them.
Spreadsheets drift
Quote spreadsheets start simple and grow. CloudNC's article lists the usual problems: outdated versions, stale material prices, assumptions buried in cells and similar jobs priced inconsistently.
How AI solves it
The aim is a pipeline that turns an RFQ email into a pre-filled quote sheet with the right questions already flagged. A person still sets the price. Here is a workflow that suits most job shops.
1. Capture the RFQ
A shared quoting inbox (or a form on your website) triggers a workflow in a tool like n8n or Power Automate. It saves the email and every attachment to a job folder, unzips archives and logs the RFQ with a due date. This is plain automation. No AI is needed, and it already stops RFQs getting lost in someone's inbox.
2. Read the drawings and the email
An LLM with document and vision support reads the PDF drawings and the email together. Current models handle this well: Anthropic's documentation, for example, says Claude processes each PDF page as both extracted text and an image, so it can read title blocks, notes and tables as well as the text. It extracts a structured record: part number, revision, material and condition, quantities and price breaks, general and tightest tolerances, finish and coating, heat treat, certification or inspection requirements and the requested delivery date.
3. Check it with rules
Plain rules then check the record. Is the material on your approved list? Is there a quantity? Does the tightest tolerance exceed what your machines hold? Is there an ITAR or export marking? Anything missing or unusual becomes a flag for the estimator or a draft question back to the buyer.
4. Find similar past work
The extracted record is used to search your past quotes and jobs, by material, size, features and customer. A simple search over well-labelled history often works better than anything clever. The estimator sees the three closest past jobs with their quoted and actual hours.
5. Estimate times with the right tools
This is where you should be careful. An LLM is not a reliable way to estimate cycle times. Use your CAM software, a feature-based estimating tool or your own historical run times for that. The AI's job is to hand those tools clean inputs and to summarise risks in plain language.
6. Draft the quote, then a person prices it
The workflow fills your quote template with materials, outside processes, the suggested operations and the flags. The estimator reviews, adjusts hours, applies judgement on price and sends it. The outcome (won, lost, no quote and why) goes back into the history so the next search is better.
| Step | Best tool | Why |
|---|---|---|
| Capture and file the RFQ | Workflow automation (n8n, Power Automate, Zapier) | Fixed steps, no judgement needed |
| Read drawings, notes and email | LLM with vision and document support | Messy, unstructured inputs |
| Check required fields and limits | Rules | Must be predictable and auditable |
| Find similar past jobs | Search over your quote and job history | Accuracy depends mostly on your own data |
| Estimate setups and cycle times | CAM, estimating software, historical times | Depends on machine and process knowledge |
| Set the price and send | Your estimator | Commercial judgement stays with people |
Costs and timeline
Every figure in this section is an estimate based on the assumptions shown. Swap in your own numbers.
Assumptions: a shop receives 25 RFQs a week. Each takes 90 minutes from email to sent quote, of which about 40 minutes is preparation: opening files, reading notes, checking materials, hunting for past jobs and typing up the quote. The person doing it costs the BLS median estimator wage of $37.86 an hour before overheads.
That is roughly 37 hours of quoting a week, around 17 of them on preparation. If automation cut preparation from 40 minutes to 10 minutes of review, the shop would free about 12.5 hours a week, or roughly 54 hours a month. At $37.86 an hour that is about $2,000 a month in wage cost alone, and more once overheads are added. The bigger gain is usually speed: the same people can turn quotes around in a day instead of several, and can quote more of the work they actually want.
Model usage is small. Anthropic's documentation says a PDF page typically uses 1,500 to 3,000 text tokens plus image tokens. Assuming an RFQ pack of around five pages and 100 RFQs a month, usage would likely be in the tens of dollars a month at standard API prices.
| Option | What it involves | Best for |
|---|---|---|
| Do nothing | Keep quoting by hand from email and spreadsheets. No new spend, but the bottleneck and slow turnaround stay | Shops with a handful of RFQs a week and a quoter with spare time |
| Buy quoting software | A dedicated job shop quoting platform, usually priced per user per month, often with geometry-based estimating and a customer portal | Machining or sheet metal shops whose work fits what the product supports and who are happy to move quoting into it |
| Custom or hybrid build | An intake and extraction workflow around your existing inbox, ERP and quote template, feeding your current estimating method or a bought tool | Shops with unusual processes, mixed work (machining plus fabrication plus assembly), a legacy ERP, or a wish to keep drawings in their own environment |
Timeline (estimate): a first version covering RFQ capture, drawing extraction, rule checks and a pre-filled quote sheet usually takes four to eight weeks. Most of that time goes on agreeing what a complete RFQ looks like for your shop and testing extraction against a few hundred of your past RFQs, where you already know the right answers. Run it alongside your current process for a few weeks before relying on it.
When not to use AI
- You quote fewer than five or so jobs a week and the quoter is not stretched. A better spreadsheet and a shared inbox will do more for less money.
- A packaged quoting platform already fits your work. If you do mainly CNC machined parts from STEP files, the established products may cover most of what you need. Buy one and keep the odd jobs manual.
- Your drawings are often controlled or export-restricted. Check where data is processed and stored before sending any drawing to a cloud model, and keep restricted work out of the automated path until that is settled.
- You want the AI to set prices. Pricing depends on your capacity, the customer relationship and what the work is worth to you. Use AI to prepare and check, and leave the price with a person.
- Your past quotes have no outcomes recorded. Similar-job search only helps if you know which quotes won, which lost and what the jobs really took. Start recording that now, even before any automation.
How to start
- Pull the last 100 RFQs and time how long a typical one takes, split into preparation and actual estimating.
- Write down the fields every complete quote needs, and the reasons RFQs stall (missing material, no quantity, unclear revision).
- Set up one shared quoting inbox and a simple log of every RFQ with its outcome and a no-quote reason.
- Test extraction on 20 past RFQs where you already know the answers, and measure how often each field is right.
- Decide whether a bought platform fits your work before building anything.
If you want help designing and building the intake and extraction workflow, our AI workflow automation service covers exactly this kind of project. The same approach works on the sales side for distributors, as described in our guide to AI order entry automation, and for supplier paperwork, as in automating freight invoice audits.
FAQ
Can AI read engineering drawings for quoting?
Modern LLMs with vision can read title blocks, notes, material callouts, finish requirements and most dimensions from clean PDF drawings. They are less reliable on poor scans, dense views and geometric tolerancing symbols, so the fields that affect price should be checked by rules and a person. Treat the AI as a fast first reader whose output still gets checked.
How long should it take to quote a machining job?
It depends heavily on the part and the shop. Simple repeat parts can be quoted in minutes, while complex multi-operation parts can take hours. The useful measure is how much of that time is preparation rather than estimating, because preparation is the part automation can shrink.
Is there software that quotes CNC parts automatically?
Yes. Several job shop quoting platforms estimate machining costs from 3D models and drawings and offer customer portals. They work best for standard machined or sheet metal parts. Shops with mixed work or unusual processes often find they still need a custom intake step or manual estimating for part of their volume.
How do I reduce time spent on quotes we don't win?
Record every RFQ with its outcome and a reason when you decline to quote. Over a few months you will see which customers, part types and order sizes you rarely win, and you can respond to those faster with a simpler estimate or decline them. Automating preparation helps too, because a lost quote that took 20 minutes hurts far less than one that took two hours.
Is it safe to send customer drawings to an AI model?
It can be, if you choose a provider and set-up that meets your customers' requirements. Check the provider's data retention and training terms, where data is processed, and any contractual or export control limits on the drawings. Restricted drawings should stay out of any cloud workflow until you have confirmed it is allowed.
Sources
- eMastercam forum, "What's your quoting process?": https://www.emastercam.com/forums/topic/95750-whats-your-quoting-process/
- CloudNC, "Why quoting is the bottleneck in manufacturing" (July 2026, vendor blog): https://www.cloudnc.com/blog/why-quoting-is-the-bottleneck-in-manufacturing
- Hacker News, "Show HN: Isonq, PDF/DXF and STEP to shop quoting" (May 2026): https://news.ycombinator.com/item?id=48012600
- US Bureau of Labor Statistics, Occupational Outlook Handbook, Cost Estimators: https://www.bls.gov/ooh/business-and-financial/cost-estimators.htm
- US Bureau of Labor Statistics, Occupational Outlook Handbook, Machinists and Tool and Die Makers: https://www.bls.gov/ooh/production/machinists-and-tool-and-die-makers.htm
- Modern Machine Shop (sponsored content for Paperless Parts), "How Job Shops Can Generate Accurate Estimates in Minutes": https://www.mmsonline.com/articles/how-job-shops-can-generate-accurate-quotes-in-minutes
- Anthropic, Claude documentation, PDF support: https://platform.claude.com/docs/en/build-with-claude/pdf-support
If you would like to talk through your own quoting process and whether automating part of it is worth it, get in touch at https://sintraflow.com/contact.
