Product data that AI can read
If your catalog lists the same chair three times in three colours, an AI assistant will compare that chair with itself. Most AI shopping problems start in the product data.
What I do
- One master record per product, with variants as children.
- Required attributes per category, with nothing going live while any are missing.
- Raw and canonical values: keep what the supplier sent, show the system one clean value, and connect them with a mapping table so every change is traceable.
- Unit rules: one canonical unit per attribute, conversions in the pipeline.
- Facet governance: which filters a category gets, in what order, and the coverage threshold before a filter goes live.
Scale
From a premium Italian bag brand with around 70 SKUs, where the challenge was depth, to one catalog run across Shopify, Amazon, eBay, Fnac and Cdiscount, each with its own category tree and required fields.
Raw vs canonical
| Raw (as received) | Canonical |
|---|---|
| Black | black |
| BLK | black |
| Jet black | black |
| 12 in | 30.5 cm |