AI quoting for machine shops

Published 2026-11-03 by VisionzLab

Why quoting is the bottleneck

In a typical job shop, a few people quote everything, usually the same people who run the floor. Every RFQ has to be read, the print checked against the model, the material and stock size worked out, setups and outside processes estimated, and the missing details chased. Most quotes don't win, so most of that time goes to work the shop never makes.

Buyers have also changed. An engineer who needs a bracket tomorrow can upload a model to an instant-quote website and see a price in seconds. A local shop that takes three days to answer, or replies with a list of questions, loses that job before anyone compares quality.

Not every shop feels this. An owner who quotes a handful of jobs a week, mostly for repeat customers, may spend very little time on it. The pain sits with shops that quote often: dozens of parts a week, many of them from new buyers.

What's on the market (October 2026)

There are two different markets, and it's easy to mix them up.

Shop-side estimating tools work for the shop's estimator. They read the model and the print, estimate cycle time and cost, and build the quote. Examples: CloudNC Quote Agent, which launched in October 2026 (forward the RFQ or upload drawings and models, and it drafts a manufacturing plan, routing times and costs, and asks about anything missing), Paperless Parts, Machine Research, Toolpath, up2parts, Spanflug and DigiFabster (instant quoting on a shop's own website).

Buyer-side instant-quote platforms work for the buyer. SendCutSend, Xometry and Protolabs Network give an instant price on upload, then make the part themselves or send it to their own suppliers. For a local shop, they are competitors, not tools.

Published prices for shop-side tools, as listed by the vendors in October 2026:

• Free for manual quoting, and US$2.50 a part for an AI agent (CloudNC).
• About US$75–125 a month for small-shop quoting tools.
• US$2,000–50,000 a year for white-label or enterprise quoting.
• Many vendors don't publish prices and quote after a demo.

If a tool already fits, use it

If your requests arrive as clean STEP files and your parts are mostly milled or turned, trial one of the shop-side tools before you build anything. They are good and cheap, and they improve quickly. We'd tell a shop that, and help it set one up.

Where off-the-shelf tools fall short

The gaps we see are less about the estimating maths and more about everything around it:

• Messy input. Real requests include a phone photo of a sketch, a PDF with no model, or an email with three attachments and a revision in the body. Most tools are built around clean CAD.
• Questions asked too late. The missing tolerance, finish or quantity is usually found by the estimator, days in. The questions should go to the buyer first, before anyone prices the job.
• Your own history. A draft is only as good as the rates, machines and past jobs behind it. Without your won and lost quotes and your actual hours, any tool is guessing.
• Where files live. Several tools host in the United States only. Shops under Canada's Controlled Goods Program, or with customers who ask, may need Canadian hosting and clear deletion rules.
• How it fits the day. Quotes have to come out of your inbox, into your quote template and your ERP (JobBOSS², ProShop, E2 or a spreadsheet), not into another login nobody opens.
• Shop networks. No tool we found sits between one buyer and many local shops, cleaning up the request and routing it to the shops that fit.

How an AI quoting assistant works

The design we build is a first pass that prepares the quote, with a person approving it:

• Reads the request wherever it arrives: email, web form, PDF print, sketch photo or STEP file.
• Pulls out the spec: material, quantity, tolerances, finish, outside processes and due date.
• Measures the model with an open-source CAD kernel: bounding box, stock size, volume, material removed and basic features.
• Lists what is missing or unclear, worded as questions the buyer can answer in one reply.
• Drafts a cost from your machines, rates, materials and markups, and finds similar past jobs.
• Puts the draft, its working and its doubts in front of your estimator to approve, edit or reject.

What stays with the estimator

The assistant prepares; it never quotes on its own. The price, the delivery date, the decision not to quote, and anything it flags as uncertain stay with your people. Its account can save drafts and nothing more, so an odd request can't make it send anything.

For shop networks and associations

A regional network of shops can put the same first pass at its front door. A buyer uploads a sketch, PDF or model once. The assistant asks the buyer for what's missing, produces a clean request package and suggests the member shops that fit. Each shop still sets its own price. That makes the local route almost as easy as an instant-quote website, without a markup.

A network can also fund the build the way many associations launch member services: members pledge a founding rate, and nothing is charged unless enough shops sign up by a set date.

How to test it on your own requests

Don't judge any quoting tool, ours included, on a vendor's sample part. Run a small test on your own work:

• Pick twenty recent RFQs where you know what you quoted, whether you won, and the actual hours.
• Leave out anything under the Controlled Goods Program, ITAR or a customer NDA that forbids it.
• Measure the time to a first draft, the share of missing details it caught, and how close its cost came to your own quote.
• Decide from the numbers: buy a tool, build around your workflow, or keep quoting by hand.

Red flags

• Accuracy or savings figures before anyone has run your parts.
• No clear answer on where drawings are stored, who can see them and when they're deleted.
• Your drawings used to train the vendor's model by default.
• Quotes sent to customers without a person approving them.
• No way to use your own rates, machines and past jobs.