Industries / Industrial equipment and services / Industrial services / MRO supply
Catalogs and part data
This is the work of describing each spare part so that people and software can find it, tell it apart from similar parts and buy it again: supplier catalogs, part numbers, the item records a plant keeps in its software, and shared codes for classes of product.
4
Milestones achieved, from 1927 to 2000.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1927
An eight-page motor catalog
William W. Grainger set up an office in Chicago in 1927 to sell electric motors at wholesale, and sold mainly by mail through the MotorBook, an eight-page catalog; by 1991 its successor, the Grainger general catalog, listed more than 35,000 items with technical and application data. In its 2025 annual report Grainger said its high-touch businesses offer about 2 million products, its Zoro website about 13 million and MonotaRO, its Japanese online seller, access to about 29 million.
Credit
William W. Grainger, with his sister Margaret
1952
One item, one number
The US Defense Cataloging and Standardization Act, approved on 1 July 1952, required the Department of Defense to name, describe, classify and number every item it repeatedly bought, stocked or issued, so that "only one distinctive combination of letters or numerals or both will identify the same item" across all the services. That single number was to be used from purchase to disposal, in one catalog holding each item's description, size, weight and packing data.
Credit
US Congress (Public Law 436 of the 82nd Congress)
1964
The Big Book
MSC, founded in New York in 1941 as Sid Tools by Sidney Jacobson to sell cutting tools to machine shops, first mailed its printed catalog, "The Big Book," in 1964 at about 150 pages. It passed 3,500 pages by the mid-1990s and 4,500 by the 2019 to 2020 edition. Buyers found the part number in the book, then called or emailed an MSC sales representative; ordering online began around 2000.
Credit
MSC Industrial Supply, as told by John Gunderson in Modern Distribution Management in 2025
2000
German industry shares a classification
Large German companies, among them Siemens, BASF, Bayer, SAP and E.ON, founded the eCl@ss association, now ECLASS, to give buyers and suppliers one code for each class of product. ECLASS's own 25-year history says the first classes served procurement, and that from 2006 to 2010 it added properties and values for describing a product, such as a bearing's bore diameter, so that product records can be exchanged without remapping. ECLASS 13.0 followed in December 2022.
Credit
the founding companies of ECLASS e.V., Cologne
Watch for
One part, one record
Nestlé USA said its factories had kept the same spare part under as many as 20 different SAP material numbers (the plant's own item numbers), often without the maker's part number. After it put SPARETECH's AI tool on top of SAP, which looks for duplicates as soon as someone types a manufacturer part number and writes standard part descriptions, Nestlé said the time to find a part fell by about 50% in its own timed tests, that monthly use of the tool was above 95%, and that it can now hold some high-cost critical spares at selected factories and move them where needed, instead of stocking each one at every site.
Achieved whenNestlé or an outside party publishes before-and-after counts of duplicate records, measured search times, and the value of critical spares held across the factories involved.
Claimed by
Nestlé USA engineers Steven Gould and Andy Goldinger, at the 2025 conference of the Society for Maintenance and Reliability Professionals (SMRP), reported by Plant Services in March 2026
Open
01
Scoring part data cleanup
A public set of 1,000 storeroom item records written the way plants write them, with abbreviations, missing makers and part numbers buried in the text, each paired with a checked answer: the manufacturer, the manufacturer part number, a class code from UNSPSC or ECLASS, and which records are duplicates of each other. Then run every cleanup tool and open model that will take part against it and publish the scores.
The briefCredit · unclaimed02
From nameplate to stock item
A test of the whole step from a photo of a machine's nameplate to the right item in a plant's stock list: read the maker, type and part number from the plate, then find the matching record in a test list of 5,000 items that includes planted duplicates, or say that the part is not stocked.
The briefCredit · unclaimed
Open 01 · the brief
Scoring part data cleanup
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public set of 1,000 storeroom item records written the way plants write them, with abbreviations, missing makers and part numbers buried in the text, each paired with a checked answer: the manufacturer, the manufacturer part number, a class code from UNSPSC or ECLASS, and which records are duplicates of each other. Then run every cleanup tool and open model that will take part against it and publish the scores.
Why it is still open
The results published for AI cleanup of part records are the users' and sellers' own, such as Nestlé's 50% faster search. Researchers who label such records have not released them: a 2026 paper from Yerevan State University annotated 17,258 material and product names from one group of companies' records to train its model, and the PhRAG project for pooling spare parts (June 2026) scores its extraction on FabNER, a public set of manufacturing texts, while it collects its own spare-parts data by questionnaire. The public product sets found describe web-shop offers: WDC Products holds 11,715 offers for 2,162 products.
Built with what exists
Public product pages from distributors such as Grainger, McMaster-Carr and MonotaRO to check the true part numbers; UNSPSC or ECLASS class lists; a small model for repeated extraction, such as Meta's Muse Ember (see the Note "The jobs agents repeat"); a model that scores choices from a list, such as Google's Gemini Decision 1, to pick the class (see the Note "A thousand questions, one pass"); PhRAG's open code as one of the entrants; Python and pandas.
First step
Write 100 records by hand in storeroom style from 100 catalog items (bearings, belts, seals, fuses, filters) and have a second person check each answer.
Done looks like
1,000 records with answers checked by two people, including at least 100 planted duplicate pairs, released under an open license; scores for at least five tools or models on maker, part number, class and duplicates found, with every error listed.
Open 02 · the brief
From nameplate to stock item
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A test of the whole step from a photo of a machine's nameplate to the right item in a plant's stock list: read the maker, type and part number from the plate, then find the matching record in a test list of 5,000 items that includes planted duplicates, or say that the part is not stocked.
Why it is still open
Reading the plate has been done. The Equipment Nameplate Dataset (2019) is a public set of nameplate photos for training text readers, a 2025 paper reads motor nameplates with Google's Gemini API, and Automa.Net sells nameplate scanning. The step after reading is the one Nestlé's engineers described: technicians see the maker's part number on the tag, not the plant's material number. A search found no public test of how often a nameplate photo leads to the correct stock record.
Built with what exists
A phone camera; a vision-language model such as Gemini or Claude, or open text readers such as PaddleOCR or Tesseract; the test stock list from the deliverable above; Python.
First step
With permission, photograph 50 nameplates on machines that are out of service at a motor repair shop or a surplus dealer, and write down the correct maker and part number for each.
Done looks like
300 nameplate photos of motors, gearboxes, pumps and valves, each with its checked answer, and a published count of photos matched to the right record, matched to a wrong one, and correctly reported as not stocked.