Rank search results by what customers actually buy, not just how closely the title matches the query.
Trained on your own search clicks, carts, and orders, so ranking reflects what shoppers actually chose for a given query, not a fixed relevance formula.
Your current search engine still retrieves the candidate results, the model simply re-ranks the top results before they're shown.
Every search session becomes new training signal, so ranking quality compounds the longer the storefront runs.
Measured against held-out queries and tested live, so the improvement shows up as a conversion number, not an impression.
A short walkthrough that compares current results against re-ranked results side by side.