Designing for shopping intent, not search
Why we replaced filters and facets with a single sentence, and what broke along the way.
Marcus BellDesign Director
Traditional ecommerce search asks you to translate a human want into a machine query. You know you need 'headphones for a noisy commute under three hundred', and the interface asks you to pick a category, then a brand, then a price bracket.
That translation is where most sessions die. In our logs, 61% of users who opened a filter panel abandoned before applying a second filter.
So the composer takes the sentence directly. Intent, budget, use case and constraints are extracted in one pass, and the result set is assembled from those, with the extracted parameters shown back as editable chips so nothing feels like a black box.
What broke: discovery. Filters are bad at expressing intent but excellent at browsing, and removing them entirely flattened the number of products the average session saw. We put a structured layer back on the shop page for exactly that reason — the composer for intent, the filters for wandering.
The second lesson was about trust. People will not act on a recommendation they cannot interrogate. Every AI-surfaced product now carries the reason it was surfaced, in one line, and that single change moved add-to-cart on recommended items by eleven points.
