The ETIM classifier that cites the source for every value.
Other tools promise AI and return invented numbers. We deliver a correct ETIM class, populated features, and — next to every value — the datasheet snippet or web reference that backs it. 93 % class match and 82.6 % fill-rate against manufacturer data in real production demos.
No contract, no SaaS lock-in · Setup in under 24h
ETIM 10 has 5,640 classes and 76,625 features.
Manual classification doesn't scale. Every misclassified article kills the shop filter, hides the product in search, and breaks the B2B customer's BMECat import. Supplier data arrives incomplete, in six languages, with synonyms and abbreviations only the niche specialist understands.
Existing tools rarely tell you where a value comes from. When a reviewer asks "where does it say IP65?", nobody has the answer. That is exactly what we break.
Other tools guess — we cite the source for every value.
Other tools need weeks of template engineering — we start with your first product in under 24 hours.
Other tools are locked to one AI vendor — we swap the model the moment the next one is 10 % better.
From raw product to source-backed ETIM class — in four steps.
1. Ingest
Upload BMECat XML, CSV, Excel or datasheet PDFs — the parser extracts the relevant fields and normalizes units.
2. Classification
Gemini 3 Pro / Claude Sonnet 4.6 / GPT-5.4 propose the ETIM class. Fuzzy-class matching reroutes invalid codes to the closest matching class name. Confidence score 0-100 per product.
3. Feature filling
For every EV-code feature, we extract the value from the datasheet, manufacturer PDF or via Perplexity Sonar web search. Every numeric value ships with a ~80-character snippet as proof.
4. Validation + export
Validator filters out-of-range or hallucinated values. Backfill recovers values a validator bug rejected — without a new AI call. Export as BMECat V5.0 XML, ETIM xChange V2.0 JSON or Excel.
Real measurements, not marketing.
ETIM class match on a blind 1,000-product luminaire sample.
fill rate compared to manufacturer original data.
fill rate on leading industrial manufacturers.
1,000 products fully classified and filled.
What no other vendor has ready yet.
Hallucination shield with sources
Every numeric value carries a ~80-character snippet from datasheet or web. Invented values are filtered out by default.
PDF datasheet parsing
Pulls values automatically from manufacturer PDFs — no manual OCR, no brand-specific scripts.
Audit trail / debug log
Every per-product step traceable: prompt, model response, DB validation, feature validation. For compliance teams.
Multi-model comparison
Same input through several models in parallel — the customer sees variants and picks the best fit.
ETIM 8/9 → 10 migration
5,640 ID-stable + AI for the 74 deprecated classes. Migration as a sellable feature — most competitors don't have it ready.
Backfill without new AI calls
If a validator bug discarded values, backfill recovers them from the original run — without re-calling the model. Saves re-run costs.
How we stack up against other tools on the market.
| Other vendors | ZDS ETIM Classifier | |
|---|---|---|
| Source citation per value | No | ✓ |
| Time to first productive product | 2–6 weeks | < 24 hours |
| AI vendor lock-in | Locked to one model | Multi-provider, swappable |
| ETIM 8/9 → 10 migration | Manual | Hybrid ID-stable + AI |
| Datasheet PDF as input | Pre-process manually | Native parser |
| Per-product audit trail | Error log only | Every step traceable |
Real catalogs, measured numbers.
Lighting manufacturer
1,000-product audit, 93 % class match, ~5,900 source-backed feature values. Fair methodology, no manufacturer hint.
Energy-tech group
19,000-product catalog in queue; 150 test products successfully classified including product-vs-service distinction.
Electrical-installation mix
250+ products from leading electrical-installation manufacturers (terminal blocks, automation, switchgear) with 90–99 % fill rate — industrial gear, no surprises.
Smart-home / KNX manufacturer
20 products with marketing copy in DE/EN/ES, style brief extracted from the customer's own originals.
No SaaS lock. You pay per product, not per month.
- One-time setup, calculated individually based on catalog complexity and number of sources.
- Per product: volume-degressive pricing, no monthly commitment.
Typical volumes (10,000 products, multi-language) sit an order of magnitude below annual-subscription SaaS models like Axite / ex-AX-Semantics. Quote within 24 h on request.
Start with 100 products. Free.
Upload your 100 most critical products, we classify them in 24h with every source cited. If the quality doesn't convince you, you pay nothing. No contract, no surprises.