// DEEP DIVE | PRODUCT DATA
Taxonomy and Classification: Structure for 100,000+ Products
Product classification is the invisible infrastructure of your assortment. Where it is clean, filters work, B2B portals serve your items, and cross-selling algorithms recognize affinity. Where it is incomplete or inconsistent, products disappear from search and marketplaces reject your records. We classify along ETIM, ECLASS, or your own custom tree, AI-supported and in a fraction of the manual time.
THE PROBLEM
No standards, no trade
In B2B, classification decides whether your products show up at all. If the manufacturer says "luminaire" and the buyer searches for "lamp," the customer finds nothing. Without a shared vocabulary, the competitor's cart fills up, not yours.
Classification standards like ETIM and ECLASS are exactly that shared vocabulary. They define which class a product belongs to, which features must be maintained, and which values are admissible. They make electronic data exchange between manufacturers and distributors possible in the first place. Whoever does not serve the standards drops out of B2B marketplaces, procurement systems, and catalogs.
Manual classification does not scale. A well-trained classifier needs several minutes per article. At 100,000 articles that is a full-time team for months. On top of that, every new ETIM or ECLASS version (such as ETIM 8 to 9) forces a re-mapping. No mid-market company keeps up with that any more.
Why is ETIM or ECLASS classification necessary?
Standards like ETIM and ECLASS enable frictionless electronic data exchange between manufacturers and distributors. Without correct classification, products cannot be found, filtered, or compared in B2B portals. ETIM is the leading standard for electrical, plumbing, and tooling. ECLASS dominates in mechanical engineering, chemicals, and automotive. Both are exchanged through the BMEcat format. Anyone selling in B2B cannot avoid them.
Automatic assignment through semantics
/ OUR APPROACH
01
Step 01
Text analysis
The AI reads the existing product description, for example "screw M8 stainless steel A2 DIN 933." It extracts the relevant meaning carriers: product type screw, thread M8, material stainless steel, standard DIN 933. This works even with keyword-style or incomplete descriptions because the AI brings industry semantics with it.
02
Step 02
Class mapping
Based on the extracted features, the AI looks up the matching class in the classification tree. For our screw, in ECLASS that would be class 23-11-01-01 (hex screw). In ETIM 10.0, the matching EC class. The assignment follows the official rules of the standard. We do not interpret on our own.
03
Step 03
Feature population
Once the class is set, the related features are filled automatically: thread M8, material stainless steel A2, standard DIN 933. Where features cannot be derived unambiguously from the source text, the AI marks them as "missing" and they go into the governance pipeline.
04
Step 04
Validation and BMEcat export
Before the record goes live, a rule engine checks: are all mandatory features populated? Are values within admissible lists of values (LOVs)? Only then is the record exported BMEcat-compliant or pushed directly into your PIM. At 1 million articles, this runs overnight.
Your own shop structure, AI-mappable
Not every business needs ETIM or ECLASS. B2C shops often have their own category trees ("Women > Shoes > Sneakers"), as do specialty retailers with industry-specific logic. We map to any tree, standardized or custom. With cross-mapping (for example ETIM data for a shop that thinks internally in ECLASS), our pipeline translates between the systems automatically, so you never maintain data twice.
1 million
articles classified per night instead of per quarter
A manual classifier realistically handles 50 to 100 articles per hour, depending on complexity. For 1 million articles, a full-time team needs several months. The AI pipeline runs the same volume in one night. The critical 10 to 20 percent land in the human-in-the-loop the next morning.
The difference in numbers
Criterion
Manual
AI automation (Xanevo)
Time per article
Minutes (well-trained classifier)
Milliseconds (batch processing)
Consistency
Subjective, depends on the classifier
Objective, rule-based
Update capability
Tedious with every ETIM version jump
Automatically re-mappable
Scalability
Linear (more volume = more headcount)
Linear in compute time, not in staff
Cross-standard
Practically not feasible
ETIM to ECLASS and back via configuration
We implement standards, we do not invent them
We use the official releases from the respective standards bodies: ETIM International for ETIM (currently version 10.0), ECLASS e.V. for ECLASS, GS1 for GPC, and UNSPSC for procurement classifications. Updates to the standards are loaded into our models as soon as they are officially released. You get standards conformance, not a homegrown classification system.
Three questions classification owners ask
We re-map automatically. ETIM International provides official mapping tables between versions which we load into our pipeline. Your existing data is updated to the new version with no manual effort. Changes that affect structure significantly (new mandatory features, modified classes) get a diff view for review.
Yes. Cross-mapping between ETIM, ECLASS, UNSPSC, and custom structures is part of our setup. You maintain your data once, and the AI provides the right representation for every target channel: ETIM 10.0 for the B2B marketplace, ECLASS for the industrial customer, your own shop structure for the B2C front-end.
Very high, because standards have clear rules and the AI does not have to guess. For unambiguous articles, accuracy regularly exceeds 95 percent. Uncertain cases, such as ambiguous descriptions or special products, go automatically into the human-in-the-loop. A classifier reviews only the red cases instead of maintaining 100 percent.
Related deep dives
Data Governance
How the three-layer safety model (pre, in-process, post-validation) keeps AI classification trustworthy in production.
Ontology and Knowledge Models
What taxonomy does not cover, ontologies do: relationships between products, use cases, and properties beyond pure hierarchy.
// TAXONOMY
Let's review your classification status.
Which standards are you using today, where are the gaps, which channels demand which standard? We analyze your current inventory, identify the critical mapping issues, and deliver a concrete recommendation for your classification workflow. Result in a week.