Sorting tens of thousands of products into your own category tree by hand is impossible. But automation can do more harm than good if it is built on the wrong principles.
Rules based on words in the name are a trap
The rule “if the name contains a number and g, it’s a spinning lure” correctly sorts a few wobblers and wrongly sorts thousands of rods, feeders and leads. The word “Fluoro” in a lure name is a colour, not fluorocarbon line. “X8” is a bite alarm model, not an 8-strand braid. Errors like these are silent: the product is categorised, just wrongly, and nobody notices.
Product type, not a single word
First you need to determine what the product is (rod, feeder, lure), and only then where it belongs. Colour, model, weight and length are attributes. Material can be a type if the market divides the range that way, for example fluorocarbon lines.
Several pieces of evidence instead of one
- the supplier category (if it is uniform),
- the same EAN at reference shops and their category,
- the definition of your category: what belongs in it and what doesn’t,
- similarity to products you have already confirmed.
Categorise automatically only when at least two pieces of evidence agree. Leave the rest for review, grouped by supplier category, so that one decision settles a whole group.
Watch out for mixed groups
The supplier category “Rods” may contain both rods and spare quiver tips. If you send all of it to “Rods”, the tips end up there too. So before bulk categorisation, check that the products in the group are of the same type.
Measure accuracy
Without measuring, you don’t know whether categorisation works. Keep a verified sample of products and compare regularly. When accuracy drops, switch the automation to suggestions only.