ETIM classification: train AI on your supplier catalog

ETIM is the classification standard for technical products, used widely in the electrical, HVAC and plumbing trades. Each product gets an ETIM class, and each class has its own features: the properties that make filters, comparisons and product data exchange work. The current release, ETIM 10.0, was published in December 2024.

The standard is the easy part. The work is applying it:

• A manufacturer sends a new range, and half of it arrives without a class, or with a class from an older release.
• A supplier's description says "LED panel 60x60 4000K", and someone has to decide which class it belongs in.
• Some products sit on a border between two classes, and over the years your team has settled on which one wins. That habit lives in people's heads and in your product data, not in the standard.

Every new supplier catalog repeats the same decisions your team has already made thousands of times.

ETIM release policy.

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

Your PIM or ERP already holds:

• the product descriptions your suppliers sent;
• the ETIM class your team settled on for each;
• your decisions on the borderline cases.

That is exactly what a model needs to learn from. StayCharted AI Model Trainer doesn't learn ETIM from the outside. It learns how your team applies ETIM, using reviewed examples that conform to the applicable standard.

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 products to CSV or Excel: one row per product, with the text columns that describe it (short description, long description, supplier category, product family) and the ETIM class your team assigned.
2. Check the data before training. Data quality points out duplicate rows, classes with too few examples, and class names that look like the same class written two ways.
3. Train, then read the report. The model is tested on products it never saw during training. You see how often it was right, how it does in each class, and which classes it mixes up.
4. Classify the next catalog. Upload the supplier's file with the class column blank. You get it back with a suggested class and a confidence for every row.
5. Choose what a person checks. The report shows, for each confidence cutoff, how many rows would be flagged and how many of the model's mistakes they would catch. Rows below your cutoff go to a Review Queue, where your team confirms or corrects them. The next version of the model learns from those corrections.

Products that arrive later can be classified one at a time through the Prediction API, from your own systems.

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.

Many retailers, DIY chains and B2B marketplaces don't sell by ETIM class. They sell through their own category tree: the navigation, filters and merchandising their customers see. ETIM data arrives from suppliers, and someone maps every product into that tree.

The same approach works here, with one useful twist: the supplier's ETIM class can be one of the columns the model reads. A model trained on "description + ETIM class → our category", from the products you've already placed, learns how your tree uses the supplier's classification, and can be evaluated on cases where the ETIM class is missing or wrong.

The pattern is the same for any supplier feed: years of products already placed in your tree are the training file, and new feeds follow your conventions instead of each supplier's. More on this in e-commerce product categorization.

• Start with one product group. Begin with the codes represented in your reviewed data. Our published benchmarks go up to 100 categories in one model. For a wide range, train one model per product group, or design and validate a separate group-then-class workflow.
• It assigns the class, not the features. Filling in ETIM features (voltage, IP rating, dimensions) is a different job.
• Check classes with few examples. A category with only a few examples may be difficult to learn and evaluate. The report shows which ones.
• When a new ETIM release changes classes, relabel your history to the new release first, then retrain.
• Measure it on your own data. Every model's report is measured on your products, not on someone else's catalog.

Can AI assign ETIM classes automatically?

It can suggest them. A model trained on your own classified products suggests a class with a confidence score for each new product. Products below the confidence you choose go to a person to confirm or correct.

How many classified products 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 StayCharted read BMEcat files?

Export your product data to CSV or Excel first. The model reads the text columns you choose and writes the suggested class and its confidence into the output file.

Does the model fill in ETIM features?

No. It assigns a class. Features are filled from the supplier's data or by your team.

Is my product 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.