Classifying an industrial supplier catalog into ECLASS
How an industrial marketplace evaluated four-level ECLASS classification for supplier onboarding
Current phase: ACTIVE ENTERPRISE
An industrial marketplace needed to classify third-party supplier products into a four-level ECLASS hierarchy. The task was central to onboarding: suppliers submitted heterogeneous product data, while internal teams had to determine the correct category before attributes, compliance requirements, and downstream workflows could be applied.
The enterprise leverage Claro to classify 100,000 products across 30k+ categories using a zero-shot setup. The objective was not merely to predict a top-level category, but to select the correct branch across all four hierarchy levels and expose enough confidence and evidence for reviewers to trust the result.
Claro was assessed as the classification and validation layer above the existing supplier portal and product-data stack.
At a glance
Industry: Industrial distribution and B2B marketplace
Region: Europe
Operating model: Supplier-driven catalog with backend and customer-facing taxonomies
Scale: 100,000-product evaluation across 30k+ categories; four-level ECLASS hierarchy
Primary Claro workflow: ECLASS product classification
The challenge
Industrial classification is hierarchical. A plausible top-level category can still lead to an incorrect lower-level class, which then attaches the wrong attribute template and operational rules.
Supplier descriptions may be short, inconsistent, or written for sales rather than classification. Teams therefore had to study catalogs manually and reconcile supplier terminology with the ECLASS standard.
The business also maintained a separate customer-facing taxonomy. That created a second mapping problem: a backend standard for governance and a frontend structure for navigation and merchandising.
Why the existing approach was not enough
Flat text classification was not sufficient because it ignored the relationships between ECLASS levels. Generic tools could produce a category label without showing whether the full path was coherent.
Manual classification did not scale to continuous supplier onboarding, and a fully automatic process without review would create downstream errors.
The required design needed hierarchical candidate generation, category profiles, external evidence, confidence tiers, and a reviewer feedback loop.
The solution
Claro evaluated each product against the ECLASS hierarchy in stages, using category context and enriched product evidence to narrow the available paths.
Predictions were returned with confidence bands. High-confidence records could be considered for automated handling, while medium-confidence products entered a review queue and low-confidence records remained unresolved.
A four-person customer team validated a random sample of 1000 products. In the evaluation, classifications reached more than 70% accuracy across all four levels, providing a measurable basis for deciding where automation could be trusted and where human review remained necessary.
How the workflow works
Ingest supplier product records — Claro receives the descriptions, identifiers, available attributes, and source information supplied during onboarding.
Enrich the classification context — Additional product evidence and category profiles are used where the supplier record is insufficient.
Generate a hierarchical category path — The model evaluates the product through the ECLASS levels rather than selecting one isolated label.
Score the complete classification — Confidence reflects the quality of the proposed path and the evidence supporting it.
Route records by confidence — High-confidence products can progress, uncertain products are reviewed, and feedback is captured.
Write approved classes downstream — Validated ECLASS assignments can feed attribute templates, supplier workflows, compliance checks, and frontend taxonomy mapping.
Results and current status
The evaluation covered 100,000 products across 30k+ categories without requiring a category-specific training project first.
High-confidence cases exceeded 70% accuracy across all four ECLASS levels, while a structured validation exercise exposed where category metadata, supplier evidence, or deterministic rules should be improved.
Why this workflow matters
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.
Key takeaway
Claro made the classification decision measurable: a proposed hierarchy path, confidence, evidence, and reviewer feedback instead of an opaque category label. That creates a practical route from pilot evaluation to controlled automation.
Frequently asked questions
What is four-level ECLASS classification? It assigns a product to the correct path through all levels of the ECLASS hierarchy, not only to a broad top-level category.
Can classification work without a large labeled training set? The evaluation used a zero-shot approach supported by category profiles and external product evidence.
How is classification accuracy reviewed? Predictions are grouped by confidence and checked against customer decisions. Reviewer feedback can then improve rules and category context.
Why does confidence matter? A confidence tier lets the business automate only the classifications that meet its risk threshold and review the rest.
Can the ECLASS result trigger other workflows? Yes. The approved class can determine required attributes, supplier forms, compliance checks, and mappings into a customer-facing taxonomy.
