How to map supplier products to your own catalog categories with AI

One supplier calls an item “Hardware,” another uses “Fixings,” and your catalog needs a specific branch under Fasteners. A simple lookup works when the supplier label always means the same destination. It breaks when a broad supplier category contains several types of product.

The useful evidence is usually in the product description, material, dimensions and other attributes. Historical mappings show how your team combines that evidence to choose a category.

Use the categories your catalog actually supports. Category prediction does not resolve duplicate SKUs or invent missing dimensions.

Export a set of product rows with their final destination categories. Include the product fields that will also be supplied for new imports. Write a short labeling guide for neighboring categories: what separates a bolt from a screw in your taxonomy, or packaging tape from an electrical accessory?

Keep a stable identifier alongside each record for joining results back to the original feed, but do not rely on arbitrary IDs as predictive information. Include supplier context only where it helps interpret descriptions.

Check for near-duplicate variants. If every size of one product appears in both training and evaluation, results may overstate performance on truly new products. Also test an unfamiliar supplier or a later import batch.

If a branch has very few reviewed records, gather more examples or start at a broader category level. A model cannot reliably learn every branch from the taxonomy names alone.

1. Train on approved product-to-category mappings.
2. Inspect per-category results and errors between similar branches.
3. Publish the model version chosen for the import workflow.
4. Fill a supplier CSV or spreadsheet, retaining the original records for reconciliation.
5. Review uncertain answers and products that do not fit an existing branch.
6. Import approved mappings through your PIM or commerce system's usual process.
7. Add reviewed corrections to your examples and retrain when appropriate.

For a continuing feed, your application can call the API and write results into its staging area. StayCharted does not automatically connect to every PIM or publish listings to a store. The integration supplies those actions.

Compare the model with your current rules on the same batch. A keyword rule can be entirely adequate for a simple, stable branch.

A custom catalog tree is designed for your customers and operations. An industry taxonomy provides a shared classification language. You may need both, but they are separate output decisions.

Use reviewed examples in the standard and version you actually follow. The model does not make an organization compliant merely by producing a standard-looking code. Keep the official definitions available to reviewers.

Explore ETIM, ECLASS and the classification standards overview. For purchasing data, see UNSPSC classification.

When your taxonomy changes, record the version, update labels and test the new model before switching imports. A previous model should not silently stand in for a new category scheme.