A request for quote arrives on Tuesday morning at a 25-person machining and fabrication shop. The email is short: "Please quote the attached parts, quantities 50, 200 and 1,000, delivery in six weeks." Attached: three PDF drawings, two STEP files and a supplier quality requirements document of 30 pages. The customer is a medical device company the shop has wanted to work with for years.
The estimator is also the production planner. On Tuesday there's a machine down. On Wednesday a big order goes out. The RFQ gets looked at properly on Thursday afternoon: material from the title block, tolerances and surface finish from the notes, a thread callout that needs a special tool, anodising to be sourced from a subcontractor, cycle times estimated from memory of "something similar last year". The quote goes out on Friday. The customer replies politely that they've already placed the order.
Small manufacturers lose work like this all the time, not because their price or quality was wrong, but because the quote came too late. The expertise to price the job exists. The time to assemble everything the estimator needs doesn't.
Why the estimator is the bottleneck
Estimating well takes years of shop-floor experience, and the people who have it are usually also running production, solving problems and training others. Buyers, meanwhile, expect faster responses than ever, and often send the same RFQ to several shops. The first credible quote has an advantage, a point trade magazines like Modern Machine Shop have made for years.
Teardown: where an estimator's time goes
The two highlighted bars are almost pure information work: reading documents and searching old jobs. The routing and cycle time estimate is where the estimator's judgment is irreplaceable, but even there, a good starting point from similar parts saves time.
What hides in an RFQ
| What the estimator needs | Where it usually hides | What goes wrong if it's missed |
|---|---|---|
| Material and condition | Drawing title block, sometimes a separate spec | Wrong raw material price, wrong machining parameters |
| Tolerances and surface finish | Drawing notes and individual dimensions | A tolerance that needs grinding or a special process, not priced |
| Threads, special features | Callouts on the drawing, 3D model | Special tools or operations not included |
| Finishing and treatments | Drawing notes, spec documents | Anodising, plating or heat treatment forgotten, or wrong colour or class |
| Inspection and documentation | Quality requirements document | First article inspection, material certificates, measurement reports not priced |
| Quantities and delivery | Email text, sometimes a spreadsheet | Setup costs spread wrongly, delivery promised that production can't meet |
| Packaging and marking | Quality requirements, notes | Individual bagging, labelling or part marking not priced |
The quality requirements document is the classic trap. Thirty pages of standard text, with two requirements buried on page 17 that add real cost: a first article inspection report and material certificates for every batch.
Where AI helps
The model does the reading and the searching. The estimator does the judging.
- RFQ arrivesCustomerany timeEmail with drawings, 3D models, specifications and quality requirements.
- Extract requirementsAIminutesMaterial, tolerances, finishes, threads, quantities, delivery, inspection and documentation requirements, each with the page or drawing it came from.
- Flag the unusualAITight tolerances, special processes, requirements hidden in long documents, missing information to ask about.
- Find similar jobsAIPast parts with similar material, size and features, with their actual cycle times, scrap and costs from the ERP.
- Estimator pricesEstimator30 to 60 minChecks the extracted requirements, sets routing and cycle times, requests subcontractor prices, decides the price.
- Quote with assumptionsAI and estimatorDraft quote with price breaks, lead time and a clear list of assumptions; the estimator reviews and sends.
The step that makes the biggest difference is "find similar jobs". Most shops have years of quotes, work orders and actual times in their ERP, but finding the right comparison takes an estimator's memory and a lot of scrolling. A model that can say "three similar aluminium housings in 2024 and 2025, actual cycle times 38 to 44 minutes, scrap 2 to 4%, last quoted at €41 for 200 pieces" gives the estimator a grounded starting point in seconds.
The requirements summary
Before any pricing, the estimator gets a one-page summary like this:
RFQ from MedTec GmbH, 3 parts, quantities 50 / 200 / 1,000, delivery 6 weeks. Part A: aluminium 6082-T6, overall 120 × 80 × 35 mm, tolerance ±0.02 mm on two bores (drawing 1, dimensions 14 and 15), Ra 0.8 on bore surfaces, clear anodising class 2, part marking by laser. Part B: stainless 1.4404, M5 helicoil inserts (4). Part C: same as A, different pocket depth. From the quality document: first article inspection report in the customer's format (page 17), material certificates 3.1 per batch (page 9), individual bagging (page 22). Missing: colour of anodising for part C not stated; ask. Similar past jobs: 3 found for part A.
Every line points to where it came from, so the estimator can check it in minutes. The model doesn't set the price. It makes sure nothing that affects the price is missed.
The assumptions list protects both sides
Every quote should end with its assumptions, and the model can draft them from the summary: material as stated with 3.1 certificates, tolerances as drawn, anodising clear class 2, first article inspection included for the first batch only, packaging individually bagged, delivery six weeks from order and approved drawings. If the customer's intent differs, they say so before the order, not after the parts are made.
What it's worth
- RFQs per month
- 40
- Estimator time per RFQ today
- × 3 h
- Estimating hours per month today
- 120 h
- Estimator time per RFQ with extraction and similar jobs
- × 1.25 h
- Estimating hours per month with drafts
- 50 h
- Estimator hours freed per month
- ≈ 70 h
Seventy hours a month is almost half of an estimator's working time, returned to the person who's also your production planner. The bigger effect is on turnaround: quotes that go out the same or next day instead of at the end of the week.
Handle drawings with care
Customer drawings are confidential, and some are legally controlled: defence and dual-use parts can fall under export control rules that restrict where data may be processed and who may see it. Before connecting any AI service to your RFQ inbox, check your customers' confidentiality terms and any export control obligations, choose a provider that processes data where you're allowed to, and keep controlled drawings out of the automated flow if in doubt.
Tools that fit
Quoting software for machine shops, such as Paperless Parts or the estimating modules in shop ERPs like ProShop, JobBOSS, E2 or, in Europe, industry ERPs for contract manufacturers, handle pricing and quote documents. Some analyse 3D models to estimate machining features. The AI layer described here complements them: it reads the whole RFQ package, including the long documents nobody wants to read, extracts and flags requirements, and finds comparable past jobs in your own data.
Questions manufacturers ask
Should we quote every RFQ?
No, and a fast, polite "no quote" is better than a slow one. The model can flag RFQs that fall outside what you do (materials you don't machine, sizes beyond your machines, certifications you don't hold, quantities far outside your sweet spot) so the estimator can decide in two minutes and reply the same day. Buyers remember suppliers who say no quickly and clearly, and they come back with the jobs that fit.
Can AI read technical drawings reliably?
It reads title blocks, notes and clearly dimensioned features well, and it's improving fast. It can misread a tolerance or miss a small callout, which is why every extracted requirement links to its source and the estimator checks each one. Treat it as a very fast first reader, not as the authority.
Can it estimate cycle times?
It can suggest a range based on similar past jobs, which is often a good starting point. The routing and the final cycle time are the estimator's call, because they depend on your machines, fixtures and people.
What if our past job data is messy?
Most shops' data is. Start with what's there: part numbers, materials, quantities, quoted prices and actual hours where recorded. Even imperfect history gives better comparisons than memory alone, and it improves as you record actual times more consistently.
Will customers accept quotes with long assumption lists?
Experienced buyers prefer them. Clear assumptions show you've read the requirements and reduce disputes later. Keep them short and specific.
Rule of thumb
Let the model read every page of the RFQ and find your similar past jobs; let the estimator decide routing and price. Send the quote with its assumptions spelled out, and aim for the next day, not the end of the week.
If RFQs pile up on your estimator's desk, tell me which ERP or quoting tool you use, and I'll suggest how to prepare quotes faster without losing accuracy. The same approach helps capture shop-floor knowledge before your best machinist retires, and wholesalers use a similar reading step in order entry.
Building something with AI?
I help small businesses turn ideas into software that pays off. Tell me what you’re working on and get a free first assessment.