ECLASS classification: train AI on approved product classes

ECLASS classifies products and services across industry, from electrical components to machine parts, laboratory supplies and indirect purchases. It is hierarchical, with classes described by properties, and it keeps growing: ECLASS 16.0, released on 28 November 2025, added 995 classes across its class types (including 137 classification classes) and 1,253 properties compared with release 15.0, and is fully translated into 29 languages.

That breadth is the point of the standard, and the reason applying it is work:

• A new supplier's items arrive described in their own words, sometimes in German, sometimes in English, sometimes as a part number and three abbreviations.
• Procurement needs every purchase line in the right class for spend analysis, but the free-text order lines never say which class.
• Several classes could fit an item, and your company has long since decided which one it uses. That decision is in your master data, not in the standard.

ECLASS 16.0 release details.

Suppose you have 20,000 reviewed materials and another 2,000 to classify. That is an illustrative workflow, not a measured result: train on the reviewed items and prioritize the next batch for review.

Your ERP, PIM or spend tool already holds the item descriptions, the ECLASS class your team assigned, and every judgment call on the items that could have gone two ways. StayCharted AI Model Trainer learns how your company applies ECLASS, from those examples.

Keep your labels aligned with the standard’s definitions and release. A model can repeat past mistakes as well as useful patterns; review the history before training.

1. Export your classified items to CSV or Excel: one row per item, with the text columns that describe it (short text, long text, manufacturer, material group) and the ECLASS class you assigned. A model can read several columns together.
2. Check the data before training. Data quality finds duplicate rows, classes with too few examples, and class names written two ways.
3. Train, then read the report. Accuracy is measured on items the model never saw during training, with accuracy per class and the classes it mixes up.
4. Classify new items. Upload a file with the class column blank and get it back with a suggested class and a confidence for every row, or call the model from your own systems through the Prediction API.
5. Choose what a person checks. Pick a confidence cutoff from the report. Everything below it goes to a Review Queue, where your team confirms or corrects it, and corrections can become examples when you retrain.

Confidence helps prioritize review; it does not guarantee correctness. Check a sample of high-confidence answers too. Review Queue and Prediction API availability depends on your plan; see current plans. CSV/Excel export and the API are workflow options, not a native PIM or ERP integration.

An assistant can be a useful starting point when you have a few products to explore. It does not automatically know your approved labels or company conventions unless you supply that context. A trained model learns from labeled examples and gives you held-out results to inspect.

Our assistant-versus-trained-model benchmarks cover other tasks, including banking messages and contract clauses. We have not published a benchmark for this taxonomy. Those results do not establish accuracy on your product codes. Compare approaches on representative items from your own catalog before relying on them.

Industrial webshops, B2B marketplaces and retailers selling technical products usually present them in their own category tree, not in ECLASS. Supplier data arrives with an ECLASS class (or without one), and someone places every item where customers will look for it.

The same approach applies, and the supplier's ECLASS class can be one of the columns the model reads. Trained on "description + ECLASS class → our category" from the items you've already placed, the model learns how your tree relates to the standard, and can be evaluated when a supplier's class is missing or wrong. New feeds then follow your conventions, not each supplier's. More on this in e-commerce product categorization.

• Start with the part of ECLASS you use. Begin with the codes represented in your reviewed data. Our published benchmarks go up to 100 categories in one model. For a wider range, train one model per segment or main group, or design and validate a separate group-then-class workflow.
• It assigns the class, not the properties. Filling in ECLASS properties is a different job.
• Languages: the AI Classifier Model reads meaning with a multilingual pre-trained model, but our published benchmarks are in English. If your descriptions mix German and English, measure it on your own items; the report does exactly that.
• Check rare classes. A category with only a few examples may be difficult to learn and evaluate. The report shows which ones.
• When you move to a new ECLASS release, map your history to it first, then retrain.

Can AI classify materials into ECLASS automatically?

It can suggest the class. A model trained on your own classified items suggests a class with a confidence score; items below the confidence you choose go to a person to confirm or correct.

Can it classify purchase order lines for spend analysis?

Yes, if you have past order lines with the class assigned. Short, abbreviated text is harder than full descriptions, so check the report's accuracy per class before relying on it.

How many examples do I need?

There is no single example count that guarantees useful accuracy. Start with representative labeled examples for each category, inspect held-out results, and add examples where categories are sparse or confused.

Does the model fill in ECLASS properties?

No. It assigns a class. Properties come from the supplier's data or your team.

Is my data used to train anyone else's model?

No. Each model is trained on your examples for your use and is not pooled with anyone else's data.