GS1 GPC brick classification: train AI on your product data

The GS1 Global Product Classification sorts products into four levels: segment, family, class and brick. The brick is the level that matters day to day: product data shared through the GDSN carries one, and some retailers use it to check and route new items. Check the published schema and the version your GDSN data pool requires before assigning codes.

Picking a brick is quick for one product. It is slow across a range:

• A brand launches 300 new SKUs and each needs a brick before its data can be published.
• A data pool or retailer receives items with a brick that doesn't match the product, and someone has to spot it.
• Some products could fit two bricks, and your team has long since decided which one it uses. That decision lives in your product data, not in the standard.

How GS1 GPC works.

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 item setup system already holds the product names and descriptions, the brick your team assigned, and every judgment call on the borderline items. StayCharted AI Model Trainer learns how your team applies GPC, 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 products to CSV or Excel: one row per product, with the columns that describe it (product name, description, brand, your internal category) and the brick you assigned. A model can read several columns together.
2. Check the data before training. Data quality finds duplicate rows, bricks with too few examples, and names written two ways.
3. Train, then read the report. Accuracy is measured on products the model never saw during training, with accuracy per brick and the bricks it mixes up.
4. Classify the next batch. Upload a file with the brick column blank and get it back with a suggested brick 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.

Checking existing data works the same way. Run products that already have a brick through the model: where the model confidently disagrees with the brick on file, it's worth a look.

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.

This is where most of the work is. A grocer, a department store or a marketplace receives supplier data with a GPC brick, and then has to place every product in its own category tree: the aisles, navigation and filters its customers use. That tree is the retailer's own convention, and no standard describes it.

The same approach works, and the supplier's brick can be one of the columns the model reads. Trained on "product name + description + brick → our category" from the products you've already placed, the model learns how your tree relates to GPC, and can be evaluated when a supplier's brick is missing or wrong.

The same model serves every supplier, so new feeds follow your conventions instead of each supplier's. If you also have product photos, StayCharted can train on pictures with the Vision add-on. More on this in e-commerce product categorization.

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.

• Start with your own range. A grocer or brand uses a fraction of all bricks. Our published benchmarks go up to 100 categories in one model. For a wide range, train one model per segment or family, or design and validate a separate family-then-brick workflow.
• It assigns the brick or category, not the attributes. Brick attributes and other product data are a different job.
• Short names are harder than descriptions. "CHOC BAR 45G" carries less than a full description. Choose useful descriptive columns and omit irrelevant identifiers.
• Check rare bricks. A category with only a few examples may be difficult to learn and evaluate. The report shows which ones.
• When GPC changes a brick you use, relabel those products, then retrain.

Can AI assign GPC bricks automatically?

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

Can it map supplier products into our own categories?

Yes. Train it on products you've already placed in your category tree. The supplier's brick can be one of the columns it reads.

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.

Can it find products with the wrong brick?

It can point to likely ones: products where the model confidently suggests a different brick from the one on file. A person decides.

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.