Three taxonomies for a $2B pet ecommerce retailer

Situation

A fast-growing online pet retailer had outgrown a category structure built when the assortment was a fraction of its eventual size. Merchants, finance and the website all used the same tree. Products were being placed by hand. Required attributes varied arbitrarily between similar products, and nobody could state the SKU count of a category without first agreeing which definition of category was in play.

The immediate symptom was findability. Customers could not reliably surface products they knew existed, and site search results were being patched product by product.

Approach

1. Audit and label. Every node in the existing tree was classified as a genuine product class, a marketing construct, or a reporting rollup.
2. Build the master product taxonomy. A 1:1 structure where each SKU has exactly one home, with a defined required-attribute set per node. This became the setup structure that drove workflow and validation.
3. Express the web taxonomy as rules. Placement rules were written as Solr queries mapping the master taxonomy and attributes onto the customer-facing 1:many navigation, so qualification was a consequence of the product’s data rather than a manual decision.
4. Split out the merchant taxonomy. A separate rollup for financial reporting, owned by the merchant organization on their own cadence.
5. Add the selling layer. An attribute-driven layer on top, so filters and discovery pages assembled themselves from attributes rather than requiring new nodes.

Running alongside this: a crowdsourced re-scrape of the entire catalog that tripled the number of populated attributes, because a rules-driven structure is only as good as the data it queries.

Outcome

• The master and web taxonomies became the backbone of site navigation and remained so through the company’s growth to over $2B in revenue and a $3.35B acquisition.
• Attribute count per product roughly tripled across the catalog, improving both organic visibility and on-site filtering.
• New SKUs began appearing in the right places on publication, without a merchandiser deciding placement.
• Financial reporting stopped depending on the website’s category structure.
• Site search relevance work became tractable, because relevance could be tuned against attributes and taxonomy rather than patched product by product.

What I would do differently

Define the required-attribute sets before publishing the master taxonomy, not alongside it. We shipped the structure first and backfilled attribute requirements, which meant a second pass through categories that were already live. The audit work is unglamorous and it is tempting to defer it. Deferring it costs more.