If you have ever typed an aftermarket part number into a supplier search and watched it return nothing — then watched the same number, with one dash moved, return seventeen results — you already know the problem. The catalogue is not wrong. The data model is.
A list pretends every number is unique
Most catalogues are, underneath, a long flat table. One row per SKU, one part number per row. The number you type either matches a row or it doesn’t. There is no concept that 03L 130 277 B and 03L130277B are the same physical thing, that 03L 130 277 J supersedes it, that Bosch 0 445 110 369 is the same injector in a Bosch box, or that VAG sold a re-manufactured variant under a third number for three years in 2014.
A graph does not pretend. A graph models the truth that thousands of numbers across OEM, aftermarket, brand-equivalent and supersession routes can all point at the same component. Every number is a node. Every "equivalent to" is an edge. The search lands on a node and walks the edges.
Worked example: Mk7 Golf MAP sensor
A Mk7 Golf 2.0 TDI rolls in with a P0106. The workshop wants the MAP/IAT sensor. They type 03L 906 051. Half the suppliers return nothing because their catalogue stores it as 03L906051. One supplier returns the wrong one because they have it filed under the petrol engine. A graph lookup walks from 03L906051 to its canonical group and surfaces:
- OEM: 03L 906 051 (VAG, current).
- OEM superseded: 03L 906 051 A (VAG, replaced 2013).
- Bosch: 0 281 002 996 (Bosch box, same sensor).
- Pierburg: 7.22184.32.0 (German aftermarket equivalent).
- Febi: 45670 (UK trade aftermarket).
- Variant fitment: CR engine codes only — filtered out the CFF/CFG petrol matches.
Six numbers, one part, one filter on engine code. The workshop sees stock at three suppliers within a five-mile radius, picks the cheapest with same-day delivery, and gets back to the job.
Worked example: Transit brake disc
A 2018 Transit Custom 2.0 TDCi needs a front disc. The DVLA record gives wheelbase and axle weight, both of which matter — Ford fitted two front disc sizes depending on GVW. A flat search by Ford 17F003 returns both, the workshop guesses wrong, the disc arrives, it doesn’t fit, the van sits for another day.
A graph search starts from the VRM, narrows to the heavy-rated variant on the DVLA record, lands on the 320mm group, and ignores the 308mm group entirely. The matches that come back — Brembo, ATE, TRW, Ferodo — are all the right diameter. The wrong part never even appears.
Where the graph comes from
A graph is only as good as its edges. Ours come from three sources: OEM technical bulletins and parts microfiches (the supersession truth), supplier-provided cross-references (the aftermarket truth) and supervised matching with manual review when confidence is low (the long tail). Every edge stores its source. Every match the API returns can show its working.
The half-of-the-time bit
We measure this. Across a sample of 12,000 part-number searches run against both a representative flat catalogue and the Parts Resolve graph over the same period, the flat catalogue returned a usable match on 48% of queries; the graph returned a usable match on 91%. The other 9% were genuinely ambiguous or genuinely unknown — and those are the ones we flag for human review rather than guessing.