What Is a Product Identity Graph?

Matching two records is a moment. A product identity graph is the memory of every such moment: the canonical product and all the identifiers, suppliers and documents that resolve to it, kept so the next supplier feed does not start from zero.

Definition

A product identity graph links a canonical product to its identifiers, suppliers and documents with scored edges, including negative edges (NOT_SAME_AS), that persist across ingestions.

Most catalog tools match and move on. A supplier file comes in, a matching step decides which rows are duplicates, the catalog is cleaned and the reasoning evaporates. Next month the same file arrives and the same rows are matched again. A product identity graph is what you get when that reasoning is kept, written down as edges between a canonical product and everything that points at it.

Record matching vs identity knowledge

So ABB AF1630101013 from a new supplier lands on the existing AF16-30-10-13 identity automatically, not because it was re-matched, but because the graph already knows the two are the same.

Positive edges

Each edge carries a source and a confidence score.

• SAME_AS: this record is the canonical product.
• VARIANT_OF: a configuration of a parent (AF16 vs AF26).
• SUPPLIED_BY: a commercial source and its SKU.
• REPLACED_BY: the manufacturer’s successor part.
• DOCUMENTED_BY: the datasheet or drawing that proves a value.

The full vocabulary and the lines between the easily confused ones are in the product relationship types guide.

Negative edges: the part most systems throw away

The graph also records what is not the product. AF16-30-10-11 is one digit away from AF16-30-10-13 but carries a 24–60 V AC / 20–60 V DC coil instead of a 100–250 V AC/DC one. A NOT_SAME_AS edge keeps that near-miss on file, so it is never silently merged in a later run and a reviewer never re-investigates a question that has been answered. Storing the rejection is the most valuable thing the graph does that a cleaned table cannot.

A minimal data model

ProductIdentity(id, canonical_name, manufacturer, mpn)
SourceRecord(id, supplier, sku, raw_text, source_file, observed_at)
Document(id, type, uri, page)

Edge(from, to, type, confidence, evidence_ref, decided_by, decided_at)
  type: SAME_AS | NOT_SAME_AS | VARIANT_OF | SUPPLIED_BY | REPLACED_BY | DOCUMENTED_BY

The details that make it work are on the edge: confidence routes the work, evidence_ref makes it defensible and decided_by records whether a rule or a person made the call.

Why it survives the next supplier feed

Because the edges are data and not a one-time computation, every ingestion inherits them. New records attach to existing identities, confirmed relationships raise confidence and only the genuinely new or ambiguous cases reach a human. That is the mechanism behind the claim that your catalog should get smarter every day: the identity graph is where “smarter” accumulates.

Common mistakes

• Merging records and deleting the originals. Keep every source record linked, so a merge can be reversed.
• Only storing positives. Without NOT_SAME_AS, near-misses return.
• Edges without evidence. An unsourced SAME_AS is a guess with a nicer name.
• Treating confidence as permanent. A supplier revision can change an edge; keep the timestamp.

Where Claro fits

A product identity graph is the identity-and-resolution core of the broader Product Identity & Evidence Graph. The entities it links are product identities, and the wider structure is described in product knowledge graph.

Related

FAQ

What is a product identity graph? +

A product identity graph links a canonical product to its identifiers, suppliers and documents with scored edges, including negative edges, that persist across supplier ingestions. It is the stored memory of identity decisions.

How is a product identity graph different from a product knowledge graph? +

A product identity graph is the identity-and-resolution core: which records are the same product. A product knowledge graph is broader and also covers attributes, compatibility, classification and other relationships between products.

What is a NOT_SAME_AS edge? +

A NOT_SAME_AS edge records that two products look alike but are different, such as two contactor variants that differ only in coil voltage. Keeping it stops the pair being merged in a later run and stops reviewers re-investigating it.

Does a product identity graph replace my PIM or ERP? +

No. It sits alongside them. Your PIM and ERP remain systems of record, and the identity graph supplies the resolved identities and evidence that are written back to them.