Industries / Mining and materials / Steel and metals / Machine shops
Quoting
Quoting is pricing a job before it is won: reading the customer's drawing or 3D model, estimating material, setups and machine time, and sending back a price and a delivery date.
6
Milestones achieved, from 1905 to 2025.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1905
A buyers' guide in print
Before a shop can quote, a buyer has to find it. Thomas' Register of American Manufacturers, sold as "The Buyers' Guide" for 1905–1906, ran to 1,290 pages listing makers by product, then by state and town. Its publishers wrote that it was arranged so a buyer could "instantly ascertain who makes any specific kind of article desired." The register later moved online as Thomasnet, and in December 2021 the online manufacturing marketplace Xometry bought its publisher; Xometry says the Thomasnet database holds more than 500,000 North American suppliers.
Credit
Thomas Publishing Company
2007
Quotes computed from 3D models
Proto Labs, founded in Minnesota in 1999 by Lawrence Lukis, launched Firstcut in the United States in 2007, a machining service quoted by software. A customer uploads a 3D CAD model, the software checks it for features "too fragile to be machined" or that cannot be machined at all, and returns an interactive quote in which changing the material, finish or quantity changes the price at once. Software then writes much of the machine programming. The company's 2012 annual report says conventional shops could take "a few days to weeks" to quote, that Proto Labs generated quotations for over 310,000 designs in 2012, and that Firstcut earned 28% of its revenue that year.
Credit
Lawrence J. Lukis; Proto Labs
2018
A runtime model retired
In 2018 Paperless Parts, then a 13-person company building an automatic quoting engine for a marketplace of job shops, had an intern, Luke Duros, predict how long a part would take to mill from its CAD model alone. He trained a straight-line (linear regression) model on run times for hundreds of parts simulated in CAMWorks, a CAM program (computer-aided manufacturing software that plans the cutting tool's path), all in 6061 aluminum with one standard set of tools. It came within a few minutes on parts under 40 minutes, a median miss of about 20%, but in use it overestimated run times "not in a predictable way," and the company withdrew the feature. Duros wrote in 2024 that the marketplace model also "proved to be untenable"; Paperless Parts now sells estimating and quoting software to job shops.
LessonA model trained on a simulator's run times learns the simulator; in Duros's words, "There is no 'correct' runtime," because each programmer and operator cuts the same part differently.
Credit
Luke Duros, Paperless Parts
2021
A factory buys a network
3D Hubs, founded in Amsterdam in 2013, gave customers an instant price and design feedback online and passed each order to one of about 240 partner factories in more than 20 countries; by January 2021 it had arranged more than 6 million parts. That month Proto Labs, which made parts in its own plants, agreed to buy it for US$280 million plus up to US$50 million more, to give its own quoting a network of outside shops for work its plants could not take.
Credit
3D Hubs (Bram de Zwart and Brian Garret, co-founders); Proto Labs
2022
A platform goes bankrupt
Fast Radius, founded in Chicago in 2017, quoted and made parts by CNC machining, 3D printing and injection molding, including at a "microfactory" in UPS's Worldport hub in Louisville, Kentucky. It listed on Nasdaq in February 2022 by merging with a blank-check company (a listed shell formed to buy a private firm), and its first annual report said there was "substantial doubt" it could continue as a going concern and that the money raised would not fund its plans. On 7 November 2022 it filed for Chapter 11 bankruptcy, saying it had invested over US$200 million in its platform and served over 2,000 customers since 2020. In December a court-run auction sold its assets to SyBridge Technologies for about US$15.9 million.
LessonA software platform for quoting and making parts still had to pay for its own factories, and by its own report the money raised by listing was not enough to do so.
Credit
a failure; Fast Radius, led by co-founder Lou Rassey
2025
Instant quotes at scale
Xometry, founded in 2013 and based in Maryland, runs a marketplace where a buyer uploads a part, gets a price from software trained on past quotes and orders, and Xometry then places the job with one of its supplier shops at a lower price. For 2025 it reported 81,821 active buyers, 4,996 active suppliers, US$629.6 million of marketplace revenue and a net loss of US$61.7 million, with sites in 18 languages and growth in Europe, China and Turkey. Its annual report warns that it may quote a buyer less than it then agrees to pay a supplier, and lose money on that order.
Credit
Xometry
Watch for
Quotes ten times faster
Uptool, a San Mateo, California, company founded in 2024, sells AI software that gathers a shop's request-for-quote emails, CAD files, drawings and bills of materials in one place and fills in estimates with built-in material and operation calculators. It says shops quote "10 times faster"; the one customer quoted, Nathan Dillon of VelocityCNC, says he "cut my quoting time 10x in 2025."
Achieved whenan outside party measures quoting hours, and how close the quoted prices came to the actual job costs, at several shops before and after they adopt it.
Claimed by
Uptool, February 2026
Draft quotes in minutes
CloudNC says its Quote Agent reads uploaded CAD files and drawings, flags machining risks and writes a draft quote that an estimator edits and approves, taking 5 to 9 minutes per request against 2 to 4 hours or more by hand. In July 2026 it said its cycle-time estimator "can produce estimates for 3+2 axis parts up to 20 times faster" (parts cut from several fixed tilted angles on a five-axis machine), and that Quote Agent was developed and tested in its own machine shop.
Achieved whena shop or lab outside CloudNC publishes the tool's estimated times and prices beside measured cycle times and final costs for a set of real jobs.
Claimed by
CloudNC, July 2026
Open
01
One part, many prices
A public test of how far quotes for the same machined part differ. Take 50 parts with open CAD files and complete drawings, in aluminum, steel, stainless steel and plastic, at quantities of 1, 10 and 100. Collect prices and lead times from instant-quote platforms in at least three regions, from AI quoting tools sold to job shops, from a general-purpose AI model given the drawing and one shop's hourly rates, and from five human estimators. Request quotes only and place no orders. Machine ten of the parts at a school shop and time them, to show whose estimates came closest.
The briefCredit · unclaimed02
Measured cycle times, in public
An open dataset of at least 300 parts that were actually machined, each with its CAD model, drawing, material, CAM program, machine model, and the setup time and cutting time measured on the machine, gathered from several shops and schools. Add a baseline estimator and publish its error, so that any quoting tool can be scored against real times.
The briefCredit · unclaimed
Open 01 · the brief
One part, many prices
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public test of how far quotes for the same machined part differ. Take 50 parts with open CAD files and complete drawings, in aluminum, steel, stainless steel and plastic, at quantities of 1, 10 and 100. Collect prices and lead times from instant-quote platforms in at least three regions, from AI quoting tools sold to job shops, from a general-purpose AI model given the drawing and one shop's hourly rates, and from five human estimators. Request quotes only and place no orders. Machine ten of the parts at a school shop and time them, to show whose estimates came closest.
Why it is still open
The side-by-side tests found are small. All3DP's article "We Compared Prices From 9 Online CNC Services, Found $800+ Difference", updated in July 2025, shows its method only to paying readers. MakerVerse's July 2026 comparison covered two parts on six platforms, was run by one of the platforms and ranks itself first; two of the six returned no instant price for either part. The speed claims of AI quoting tools for job shops come from their makers. Searches in November 2026 found no independent test across many parts, and a GitHub search for "cnc quote" found only small one-person quoting apps.
Built with what exists
The ten STEP models in Schmitz and colleagues' CC BY benchmark on Zenodo as a start, plus parts from other open CAD libraries; FreeCAD to draw and tolerance them; each platform's own quote page; an AI model such as Claude Fable 5.2 (see the Note "Three working devices"); a spreadsheet; volunteer estimators from local machine shops.
First step
Add tolerances, material and finish to drawings of the ten Zenodo parts and request quotes for them on three platforms in one week.
Done looks like
50 parts at three quantities quoted by at least eight sources, with every price, lead time and binding or non-binding status published, the spread for each part reported as highest price over lowest, and the same parts quoted again six months later.
Open 02 · the brief
Measured cycle times, in public
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open dataset of at least 300 parts that were actually machined, each with its CAD model, drawing, material, CAM program, machine model, and the setup time and cutting time measured on the machine, gathered from several shops and schools. Add a baseline estimator and publish its error, so that any quoting tool can be scored against real times.
Why it is still open
Paperless Parts trained its run-time model on simulated times and withdrew it. The Zenodo benchmark by Marius Schmitz and colleagues (September 2025) gives ten STEP files and their G-code but no measured times. Arıkan and colleagues (arXiv, August 2025) predicted cost from 13,684 drawings of automotive parts with about 10% average error, and the paper's page does not say the data are released. A GitHub search for "machining time" found one undergraduate thesis repository and a calculator, and no dataset of measured times.
Built with what exists
The open MTConnect agent (mtconnect/cppagent on GitHub) or each machine's own cycle timer to log start and stop times; FreeCAD or Autodesk Fusion for the CAM programs; Python with scikit-learn for the baseline; Zenodo to publish under CC BY.
First step
Ask a technical college's machining program to log the setup and cycle time of every student job for one term and to keep the CAD files.
Done looks like
300 parts from at least three shops and five machine models, with times taken from machine logs, published under an open license; the baseline's median error reported on parts from a shop it was not trained on.