Blog · · 5 min read

Faster quotes for job shops: what AI can read from an RFQ, and what still needs a person

An RFQ lands with drawings, a 3D model and a deadline. How AI can read the package, flag what is missing, check manufacturability and draft the questions back, and why the price stays with the estimator.

A drawing, a 3D model, a quantity and a deadline arrive in the inbox. Before anyone can name a price, someone has to read all of it: the notes in the title block, the one tolerance that changes the process, the coating that has to go out to a partner. In many job shops this reading, not the machining, is what slows the answer down.

This article is about that first step on the supplier side: what AI can take over when an RFQ comes in, and where an estimator still has to decide.

What a quote really depends on

A price is the sum of many small decisions. An estimator works through a list like this, often in their head:

  • Material and stock: the grade, whether it is in stock, and the size of the blank.
  • Processes and setups: milling, turning, grinding, wire cutting, and how often the part has to be clamped again.
  • Critical features: the tightest tolerances, surface finishes, threads, and inside corners that need a small tool.
  • Outside work: heat treatment, anodizing, plating or painting, usually quoted by a partner.
  • Inspection: a simple check, a measurement report, or a first article inspection.
  • Quantity and lead time: price breaks, batch sizes, and whether the date is realistic.

None of this is difficult. It is slow because the information is spread over a PDF drawing, a 3D model, an email and sometimes a spreadsheet, and because something is almost always missing.

Where the hours go

Ask an estimator where the time goes, and the answers sound alike:

  1. Reading and re-typing. Material, finish, quantities and notes are copied from the drawing into the shop's own sheet.
  2. Finding the one feature that changes everything. A tolerance of a few microns on one bore can turn a simple milled part into a ground one.
  3. Chasing missing information. No material grade, a thread without depth, a drawing that does not match the model.
  4. Waiting for partners. The coating or heat treatment quote often takes longer than the machining estimate.

The first three are reading and checking. That is where AI helps today.

What AI can do with an RFQ

Read the package into a structured list. AI can pull material, quantities, finishes, tolerances, thread callouts and notes out of drawings, PDFs and emails and put them into one table, with the page or view each value came from. The estimator checks the list instead of building it.

Compare drawing and model. A hole that is 6 mm in the drawing and 6.5 mm in the 3D model is a question for the customer before it becomes a scrapped part. AI can compare the two and list every difference.

Flag what is missing or contradictory. A material without a grade, a surface finish that the process cannot reach, a general tolerance note that conflicts with a dimension: these become a short list of questions, written in plain language and ready to send.

Check manufacturability. Deep narrow pockets, sharp inside corners, thin walls and features that need another setup can be found in the model geometry. We described these rules for different processes in 3D print, CNC or mold?.

Estimate a first range. From the features, the material and the setups, AI can give a rough range for machining time, with every assumption written down. A range with its assumptions is useful. A single number without them is not, and we explained why in Can you trust AI with engineering numbers?

Draft the reply. The structured list, the questions and the first estimate can go into a draft quote that the estimator finishes.

On makemachine.ai, the same reading and checking runs on the buyer side, when the platform prepares a package for quotation: drawings with the tolerances that matter, a solid model, quantities, materials and finishes. What goes into a good package is in AI for RFQs. The better the package, the faster any supplier can answer it.

What stays with the estimator

The price itself. It depends on things no drawing contains: how busy the machines are, how much the shop wants this customer, what a similar job really cost last time, and how much risk sits in one tight tolerance. AI can prepare the numbers behind the price, but the decision belongs to the person who answers for it.

The same is true for promises. A lead time is a commitment of the whole shop, not a calculation.

How to start without risk

You do not need to change how you quote to find out whether AI helps.

  1. Pick one part family. For example, milled aluminium housings or turned shafts.
  2. Measure today. How long does a typical quote take, from email to sent offer?
  3. Run AI alongside. For a few weeks, let it read the same RFQs and prepare the list, the questions and a first range, while the estimator quotes as usual.
  4. Compare. Where did it save time, where did it miss something, and how close was its range to the final price?

After a few weeks you know whether it pays off for your parts, based on your own numbers.

A note for manufacturers in Singapore

Singapore's Catalytic AI programme names automated preparation and processing of RFQ documents as one of the use cases it supports, with up to 70% funding for approved solutions. Our guide to Catalytic AI sums up what is known so far and how to prepare.

The short version

  • In many job shops, reading an RFQ takes longer than estimating it.
  • AI can turn drawings, models and emails into one checked list, with the source of every value.
  • It finds missing information and contradictions and drafts the questions back.
  • It checks the geometry for manufacturability and gives a first time range with its assumptions.
  • The price, the lead time and the risk stay with the estimator.
  • Start with one part family and compare against your own quotes.

Watch hardware get built by AI. Follow public projects on makemachine.ai, read every result in full and get an email when there is something new.

Get started See the live projects

More from the blog

Open-hardware projects built with AI