Search teams are rebuilding indexes for intent, not keywords — occasion, constraint, and silhouette travel together.
Model updates help, but attribution and size availability still decide whether a result feels trustworthy.
We watched a luxury team A/B test semantic search against classic filters. Conversion moved when explanations appeared.
Reshaping search is half ML, half merchandising vocabulary.
The update that matters is the one your stylists believe.
Retail search is changing because people do not search for fashion like they search for cables.
They search in fragments of life: “office dress but comfortable,” “shoes for walking in Paris,” “jacket like this but warmer,” “wedding guest outfit that does not look bridal.” The model update matters only if it helps those fragments reach products that feel right.
A luxury team testing semantic search discovered that the result list improved when explanations appeared. Customers did not only want products. They wanted to know why the product matched the intent. “Works for evening because of the satin finish and narrow silhouette” creates more confidence than a silent grid.
But model improvements did not solve everything. Size availability still decided trust. Product attributes still needed cleaning. Some categories had rich editorial tags; others were described like warehouse objects. The model could understand intent, but the catalog did not always have enough language to answer.
Stylists became part of the evaluation loop. They reviewed queries that performed well statistically but felt wrong aesthetically. They taught the team that “minimal” can mean different things across brands, price points, and markets. They corrected examples where the search engine found similar products but missed the social situation.
The lesson is that search quality is now half machine understanding and half merchandising vocabulary. A better model can open the door. The brand still has to furnish the room.
