A product taxonomy is the way you organize everything you sell. When it is done well, customers find what they want in two clicks, filters return sensible results, and adding a new product takes minutes. When it is done poorly, every new product creates a new category, filters show empty results, and your team argues about where things belong. This guide shows how to build a taxonomy that scales from 50 products to 50,000.
Categories vs. attributes vs. variants
These three concepts are often mixed, and mixing them is the root of most taxonomy problems.
-
Category answers where the product lives in your store's navigation. A product should have one primary category, even if it appears in multiple collections. Example:
Apparel > T-Shirts. -
Attribute answers what the product is, in a structured way. Attributes are facts like material, brand, fit, pattern or origin. They power filters and search.
-
Variant answers which one the customer wants when the product comes in multiple forms. Size and color are variants; material is usually not. If the customer must choose it before adding to cart, it is a variant. If it just describes the product, it is an attribute.
Keeping these distinct prevents the most common mistake: turning every attribute into a category. "Red T-Shirts" should not be a category. "T-Shirts" is the category; "Color = Red" is the attribute or variant.
Design the category tree first
A good category tree has three qualities: it is shallow, it is consistent, and it leaves room to grow.
Shallow means no more than three levels. Apparel > T-Shirts > Graphic Tees is navigable. Apparel > Men > Tops > T-Shirts > Graphic > Licensed > Marvel is not. Every extra level halves the number of products that reach it and doubles the chance a product is filed in the wrong place.
Consistent means every branch follows the same logic. Do not organize one branch by who it is for (Men, Women) and another by what it is (T-Shirts, Hoodies) at the same level. Pick one organizing principle per level.
Room to grow means leaving empty branches intentionally. If you sell only t-shirts today but plan to add hoodies, create Apparel > T-Shirts and Apparel > Hoodies now, even though one is empty. It is easier to fill an empty category later than to restructure a flat list of 500 products.
Attributes that make filters useful
Filters fail for two reasons: too many and too few. Too many attributes and the sidebar becomes a wall of checkboxes. Too few and customers cannot narrow results.
Start with five to seven attributes that matter for purchase decisions in your vertical. For apparel, that is often Brand, Size, Color, Material, Fit and Price Range. For food delivery, it is Category, Dietary Label, Spice Level and Price. Define the allowed values for each attribute in advance — a controlled vocabulary. "Red", "Crimson" and "#FF0000" should not be three different colors.
Map vendor data to this vocabulary on import. Vendors will send Cotton, 100% Cotton and Cotton Blend for the same thing. Your import map should normalize them to one value. CSV Pilot's category and tag normalization does exactly this: you describe the map once, and every new vendor file is translated to your vocabulary.
Variants: the expensive part to get wrong
Variants are expensive because they multiply. A t-shirt in 4 sizes and 3 colors is 12 SKUs, 12 inventory records and 12 rows in a Shopify export. The cost is not just data entry — it is every future price change, inventory update and image assignment multiplied by 12.
Rules of thumb:
- Only make a variant if the customer must choose it to buy. "Size" is a variant. "Weight" for a 500g pack is not — it is a separate product.
- Keep the number of option dimensions to two. Size and color is manageable. Size, color and material (4 × 3 × 3 = 36 variants) is rarely worth the complexity.
- Use the delta pricing model: the product's base price covers the default variant, and each other variant stores only the difference. This keeps bulk price changes simple and is how CSV Pilot exports variants.
Governance as the catalog grows
A taxonomy without governance drifts. After a few months, you will find T-Shirts, T-shirts, Tees and T Shirt as four categories. Prevent this with three habits:
- One owner. One person approves new categories and attribute values.
- A change log. Every new category documents why it was needed, with an example product.
- A quarterly review. Run a report of categories by product count. Any category with one product is a candidate for merging.
A clean taxonomy is invisible to customers — products are just where they expect them to be — but it is the foundation that makes every future CSV import, filter and report work without manual cleanup.