Where things stood
The marketplace needed to classify 100,000 supplier products across more than 30,000 ECLASS categories while keeping accuracy across all four hierarchy levels. Supplier descriptions were inconsistent and written to sell, not to classify, and manual classification could not scale with continuous onboarding.
What changed
Claro ran a staged hierarchical evaluation using category context and enriched product evidence, generating complete category paths with confidence bands — high-confidence records routed to automation, medium-confidence to human review, and low-confidence to manual resolution.
What it looks like today
Zero-shot classification across 100,000 products, with no category-specific training required.
Over 70% accuracy across all four ECLASS levels in high-confidence cases.
A structured validation pass that surfaced where category metadata and supplier evidence needed improvement.
Classification is not an isolated tagging task. In an industrial catalog, the chosen class determines attributes, review rules, compliance expectations, and how suppliers enter the ecosystem.