Industries / Retail, hospitality and personal services / Retail and e-commerce / Auto parts retail
Catalogs and fit
Catalog work records which part fits which vehicle, so that a counter clerk, a mechanic or an online shopper can pick the right one from millions of parts.
5
Milestones achieved, from 1984 to 2020.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1984
Catalogs move onto screens
Until the 1980s a parts counter held racks of paper catalogs, one for each supplier, reprinted every year or so with supplements in between, according to Courtney Pedler's 2025 history of parts catalogs in Aftermarket Business World. Triad Systems launched an electronic catalog on 15 August 1984 with about 30,000 part numbers for a few hundred thousand vehicle applications, and by the late 1980s thousands of distributors and independent parts stores subscribed to its monthly release, by the account of Activant, Triad's successor, which says Triad invented the aftermarket electronic catalog.
LessonEach electronic catalog company built its own vehicle table and file format, so parts makers had to convert one catalog into many, at high cost.
Credit
Triad Systems Corporation (later Activant, now Epicor)
1994
A shared format in Europe
In 1994, 21 companies from the international spare parts trade set up TecDoc, a common format in which each vehicle type is defined once and linked to the parts that fit it. TecAlliance, which now runs it, says that before the data was standardized, matching a part to a vehicle type relied on trial and error and gave high error rates.
Credit
the 21 founding parts companies; TecAlliance
2003
North America agrees a standard
In the late 1990s a parts maker could need up to five versions of its digital catalog, one for each trading partner, and each named the same 1986 Chevrolet half-ton pickup differently, Pedler writes. The trade association, then the AAIA and now the Auto Care Association, issued a first standard, a flat table of makes, models and years, in 1997, and replaced it from 2003 with ACES (Aftermarket Catalog Exchange Standard), developed from a vehicle database donated by the repair-software firm Mitchell. PIES, a standard for product details, followed. By 2005 Activant estimated that 80% to 85% of parts transactions came out of electronic catalogs.
LessonA standard needs good data behind it: an electronic catalog developer told Auto Service World in 2005 that early users of the 1997 standard "may have had a bad experience" because makers' data was unreliable.
Credit
Scott Luckett and volunteers at the AAIA; Will South of Dana, for the catalog managers' group; Technologue and Mitchell, which donated vehicle tables
2011
Part numbers by VIN
A European Commission regulation of 8 June 2011 required car makers to give independent operators, including parts makers and sellers, a database that links each vehicle identification number (VIN, the code that identifies one vehicle) to its original-equipment part numbers and part names, with the dates each part is valid from and to. The database must be updated regularly, including changes made to a vehicle after it was built.
Credit
European Commission (Regulation (EU) No 566/2011, amending Regulation (EC) No 715/2007)
2020
Data a computer can read
Kia gave independent operators read-only access to its repair database, and Germany's parts-trade association, Gesamtverband Autoteile-Handel, sued for the data in a form its members could process. On 19 September 2019 the EU Court of Justice ruled that the 2007 law did not require information "in a form that is amenable to onward electronic processing." The court noted that this duty was added only by Regulation (EU) 2018/858, which applies from 1 September 2020 and requires the VIN-to-part data as "machine readable and electronically processable datasets."
LessonA right to look data up on a screen did not let a seller load it into its own catalog; that took a new law.
Credit
Court of Justice of the EU (Case C-527/18); European Parliament and Council (Regulation (EU) 2018/858)
Watch for
AI-built fitment records
Motormia, a parts discovery app founded in Arizona in 2023, says it has used AI to build a dataset of more than 5 million aftermarket parts from over 4,000 brands, mapped to more than 75,000 vehicle configurations in over 50 million fitment records. The announcement gives no error rate and no method for checking the records.
Achieved whenan outside party checks a random sample of the records against parts makers' own catalogs and publishes the error rate.
Claimed by
Motormia, May 2026
Open
01
An open, checked fitment set
A free table of 2,000 part-to-vehicle pairs for five everyday parts (oil filters, cabin air filters, wiper blades, front brake pads and batteries) on 100 vehicles common in North America and Europe. For each pair it records what at least three parts makers' own public catalogs say, with the page and date of each reading, and flags where they disagree.
The briefCredit · unclaimed02
AI answers on fitment, scored
A test of 300 fitment questions, each asking which part, such as an oil filter, fits a given vehicle and engine, with answer keys taken from two parts makers' catalogs. It is put to the main AI models with and without web search, and scores right answers and confident wrong part numbers separately.
The briefCredit · unclaimed
Open 01 · the brief
An open, checked fitment set
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A free table of 2,000 part-to-vehicle pairs for five everyday parts (oil filters, cabin air filters, wiper blades, front brake pads and batteries) on 100 vehicles common in North America and Europe. For each pair it records what at least three parts makers' own public catalogs say, with the page and date of each reading, and flags where they disagree.
Why it is still open
The reference databases that ACES fitment files depend on require "an annual paid subscription," and the 10,000 eBay listings with known fitment used in eBay's 2024 student machine learning contest were given to entrants to build models, not published as a test set. A search found no public set for checking fitment claims.
Built with what exists
NHTSA's free vPIC VIN decoder and its downloadable database, for the vehicle list; parts makers' public catalog lookups, read by hand or with a browser agent such as Browser Use where a site's terms allow it (see the Note "Deleting 80% to start over"); Python; CSV files on GitHub.
First step
Take 20 common vehicles and look up the oil filter in three makers' online catalogs, recording each answer and its link.
Done looks like
2,000 pairs, each with at least three catalog readings; a second person rechecking 200 random readings with at least 95% agreement; and a published rate of disagreement between makers for each part type.
Open 02 · the brief
AI answers on fitment, scored
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A test of 300 fitment questions, each asking which part, such as an oil filter, fits a given vehicle and engine, with answer keys taken from two parts makers' catalogs. It is put to the main AI models with and without web search, and scores right answers and confident wrong part numbers separately.
Why it is still open
The nearest published test covers industrial parts, not vehicles. In Reshape Automation's index of October 2026, three AI models got 12% to 14% of 100 questions right from memory and the best got 52% with web search; about one in three wrong answers gave a specific part number or figure with no caveat. Reshape wrote the questions and the answer key, and no third party reviewed them. A search found no such test for vehicle fitment.
Built with what exists
The checked fitment set above as the answer key; the model makers' APIs; an open evaluation tool such as the UK AI Security Institute's Inspect; a second person to grade disputed answers.
First step
Write 50 questions from one filter maker's catalog and run them on one model.
Done looks like
300 questions run three times on each of at least three models, with accuracy, the share of wrong answers that name a specific part number without a caveat, and every question and answer key published.