Your product descriptions
Use the text that describes the item: its name, material, use, product family and relevant supplier information.
← All solutions
CLASSIFY TO INDUSTRY STANDARDS
ETIM, ECLASS, GS1 GPC and HS code classification
Classify new products using the decisions your team has already reviewed. Train StayCharted AI Model Trainer on your approved codes, get suggestions with confidence scores, and prioritize what people check.
SUPPLIER DATA → YOUR CATEGORY
Supplier row: LED panel 600x600 4000K 36W, with the supplier’s ETIM classCeiling Lights (your category)Illustrative category mapping · Not a measured prediction or an ETIM class assignment
You’ve already done the hard part.
Your PIM, ERP or customs records contain supplier descriptions, reviewed labels and decisions on borderline products. Those examples can become a training file for your next catalog.
Your reviewed labels
Choose the target: an ETIM class, ECLASS class, GPC brick, HS code or your own retail category. Keep definitions and releases consistent.
A report before reliance
Measure performance on held-out examples, check errors by class and decide how predictions will be reviewed.
For example, a team with 20,000 reviewed products could use them to prepare suggestions for its next 2,000. This illustrates the workflow; it is not a measured result or an accuracy promise.
Choose your classification standard
| Standard | Who it is for | The work | Guide |
|---|---|---|---|
| ETIM | Electrical, HVAC and plumbing wholesalers and manufacturers | Supplier ranges without classes, older releases and borderline products | Read the ETIM guide → |
| ECLASS | Manufacturers, industrial distributors and procurement teams | Abbreviated material descriptions and purchase lines without an approved class | Read the ECLASS guide → |
| GS1 GPC | Brand owners, data pools, grocers and retailers | Brick codes for new SKUs and checking existing product labels | Read the GS1 GPC guide → |
| HS codes | Importers, exporters and cross-border sellers | Preparing code suggestions from reviewed history for compliance teams or brokers | Read the HS codes guide → |
From supplier description to your classification
Export reviewed products
Use CSV or Excel with one row per product, the descriptive columns, and the approved class, brick or code. Use a consistent standard release and retain codes as text when needed to preserve leading zeros.
Check your training data
Review duplicate rows, sparse categories and inconsistent label names. Confirm that past assignments follow the applicable definitions; a model can learn past mistakes too.
Train and inspect the report
See results on examples held out of training, including accuracy by category and the classes the model confuses. Check the categories that matter to your team.
Classify the next supplier file
Upload a file with the target column blank to get suggested categories and confidence scores, or integrate predictions through the API on a supported plan.
Review, correct and improve
Choose a confidence cutoff and review flagged rows in the Review Queue on a supported plan. Include approved corrections when retraining, then evaluate and publish the new version.
Confidence is a review signal, not a guarantee. Sample high-confidence answers too, and check current plans for Review Queue and Prediction API access.
RETAILERS & MARKETPLACES
Your category tree, across supplier feeds.
You may receive standard codes from suppliers but sell through your own aisles, navigation and merchandising categories. Train on products your team has already placed in that tree.
The supplier’s code can be an input alongside the description. Measure performance separately when codes are missing or wrong; neither condition is automatically solved by training.
Two different targets
Assign a standard code: descriptions → reviewed ETIM, ECLASS, GPC or HS labels.
Map to your retail tree: descriptions + supplier code → your own categories.
Why train a model instead of just asking an assistant?
An assistant can be a useful starting point for a few products. It does not automatically know your approved labels or supplier shorthand unless you give it that context. A trained model learns from your examples and lets you evaluate how well it reproduces your classifications.
Our published benchmarks cover other tasks, including banking messages and contract clauses. We have not published ETIM, ECLASS, GPC or HS accuracy results. Test your own products and category boundaries before choosing an approach.
What to expect before you start
- Start with the part of the standard you use. Published benchmarks cover up to 100 categories in one model. For a broader range, evaluate separate models by product group or design and validate a group-then-class workflow.
- Classes, not attribute enrichment. Filling ETIM features, ECLASS properties or GPC attributes is a separate job.
- Inspect rare classes. Sparse examples can make a class difficult to learn and evaluate. Add representative labels where needed.
- Review release changes. Update and validate your labels against the applicable standard before retraining.
Common questions
Can AI classify products into ETIM, ECLASS, GPC or HS codes automatically?
It can suggest labels learned from your reviewed examples, with confidence scores to help prioritize human review. Confidence does not establish correctness. HS suggestions require a classification decision by your compliance team or broker.
How many classified products do I need?
There is no single count that guarantees useful accuracy. Start with representative examples for the categories you use, inspect held-out results, and add examples where classes are sparse or confused. Our published benchmarks cover up to 100 categories in a model; larger taxonomy workloads need their own evaluation.
Can it map supplier products into our own categories?
Yes. Train on products already placed in your own category tree. The supplier’s ETIM class, ECLASS code or GPC brick can be an input column alongside the description. Evaluate cases with missing or incorrect supplier codes separately.
Does it work with BMEcat or GDSN files?
Export the relevant product data to CSV or Excel first. StayCharted reads the text columns you select and returns suggested labels and confidence scores. This workflow does not claim a native BMEcat, GDSN, PIM or ERP connector.
Does it fill ETIM features, ECLASS properties or GPC attributes?
No. This workflow assigns a class, brick or code. Attribute enrichment is a separate task using supplier information and your team’s review.
Is my product data used to train anyone else’s model?
No. Your models learn from your examples for your workspace. Your uploaded data is not pooled to train other customers’ models.
Explore related workflows
Start with a product group you know.
Bring reviewed examples, inspect the results and decide what your team needs to check.