What Is Product Classification?
Product classification assigns each product to the appropriate category or family in the catalog taxonomy.
Classification is the operational step that connects an individual product record to the catalog structure. It determines which category the product belongs to and, often, which attribute schema and validation rules should apply.
Classification can be performed manually, through deterministic rules or with AI-assisted matching.
See How It WorksTaxonomy defines the map; classification places the product on the map.
A taxonomy can exist without any products assigned to it. Classification is the repeated decision applied to each product or batch.
Keeping those responsibilities separate makes catalog governance easier: the structure can be controlled centrally while classification workflows evolve independently.
Three ways to assign a product to a category.
Manual review, deterministic rules and AI-assisted matching solve different parts of the same problem. Most catalogs use a combination.
AI can use several product signals instead of relying on one category label.
Title, identifiers, extracted attributes and other product context can be evaluated together to suggest the most likely target category.
The objective should be classification into an existing canonical taxonomy, not uncontrolled generation of new categories.
Evidence quality determines whether classification can continue.
ENRIVAQ does not publish one universal probability threshold for every classification. Routing depends on product identity, canonical binding, provenance and conflict state.
A clear supported assignment can continue. Competing category paths, weak identity or incomplete evidence are held for review. Invalid or wrong-product evidence is rejected.
Do not force a category when product identity or taxonomy evidence is weak.
Multiple plausible bindings or conflicting evidence stay unresolved.
Clear identity, canonical target and supporting evidence allow the assignment to proceed.
Classify from product evidence, not from a vague supplier label.
The AC820825 reference arrives as a Kverneland fan-impeller record. Identity signals, compatible machine context and the product type support a much more specific canonical placement than a generic spare-parts bucket.
The classification outcome determines which attribute family applies next. That matters because a fan impeller should not inherit the schema of a bearing, blade or generic “tractor part”.
Classification problems often come from unclear taxonomy rather than the classifier itself.
Duplicate targets, generic families, title-only decisions and forced assignment create structural errors that propagate downstream.
Common questions about product classification.
Automation, taxonomy ownership, useful input data and the link between classification and enrichment.
Automation still needs an ambiguity path.
Some records can be classified automatically, but ambiguous technical products may still require review.
PARTIALLYSee how classification routes products into the right family and schema.
Classification is not an isolated label. It determines which family, attribute schema and validation logic become relevant to the product record.