Our summary
A shop may group its products into departments and smaller categories. This recipe compares two ways to ask Jev where a product belongs: show every final category together, or first narrow the list to a few promising departments.
The staged version keeps the best two departments rather than committing to one immediately. It then asks Jev to choose among their smaller categories. Keeping alternatives gives products near a boundary another chance, while shorter option lists can reduce the amount of text sent.
Both methods worked on the author's small set of clear product titles, which does not establish performance on difficult stock. Staging uses more requests and can still discard the right branch. Category size, description length, and actual mistakes should guide which approach a team uses.
Key takeaways
- Try a single choice when the category list is manageable.
- Keep more than one possible parent category for boundary cases.
- Compare request count and mistakes, not only the amount of text processed.