Building a taxonomy that matches how customers search
A taxonomy is your answer to "how does this catalog divide into categories?" It sounds abstract until you realize you've already got one - it's implied by your collections, your navigation, and your product types. The question is whether it was designed, or whether it accumulated.
Internal logic vs. customer language
Taxonomies fail the same way navigation does: they mirror the company instead of the customer. Categories named after product lines ("The Studio Collection"), supplier groupings, or internal codes. The customer looking for a linen shirt doesn't know your product lines. They know "shirts," "linen," maybe "summer."
The test for every category: would a first-time visitor know what's inside from the name alone? If it needs explaining, it's internal language.
Read the demand, don't guess it
You don't have to imagine how customers think - the data exists. Three sources, in order of usefulness:
Your own site search. What people type into your store's search box is your customers' taxonomy in their own words. If "blazer" is searched constantly and doesn't exist as a category, that's a category waiting.
Search Console. The queries your store already appears for reveal the language of demand at large - including subcategories you'd never have invented. (Phase 3 covers reading this properly.)
Your filters. Which filters do customers actually use? Heavily-used filter values - a material, a fit - are often categories in disguise, worth promoting to real collection pages.
Depth and the "roughly ten" rule
Two levels serve most catalogs: top categories (Shirts, Trousers, Dresses) and one level of subdivision (Linen shirts, Overshirts). A useful heuristic: a category earns its existence when it would contain roughly ten or more products and represents real search demand. Below that, it's a filter, not a category. A taxonomy of fifty micro-categories with three products each is worse than eight strong ones - for browsing, for ranking, and for maintenance.
Write it down
The output of this work is one document: your category tree, with a one-line definition of what belongs in each. That document then drives collections, navigation, product types, and feed categorization - which is the whole point. One taxonomy, applied everywhere, instead of four systems that almost agree.