Recommendations

upstart commerce provides advanced recommendation services to enhance product discovery and improve customer engagement these services leverage ai driven models to offer personalized suggestions, ultimately increasing catalog exposure and driving conversions the recommendation services are categorized into three primary types visual recommendation – provides recommendations based on product image similarity alternate product recommendation – suggests alternatives based on product attributes get recommended items – uses machine learning and user feedback to provide highly relevant personalized recommendations both services are seamlessly integrated into upstart commerce search, ensuring that businesses can maximize product visibility while enhancing the shopping experience architecture overview frontend interaction user interaction – customers viewing a product receive relevant recommendations recommendation retrieval – the system fetches recommendations based on either visual attributes or product features data handling & processing catalog updates – ensures recommendations stay updated as new products are added embedding generation – converts product images into numerical representations for comparison in the visual recommendation service attribute based matching – uses product details like title, description, tags, material, and brand in the alternate product recommendation service storage & retrieval – stores embeddings and product attributes for retrieval automated recommendations updation – ensures recommendation accuracy as catalogs are updated query rule filtering – businesses can apply brand specific filters example use case apparel store a customer searching for a "summer dress" selects a floral dress the system then recommends other floral dresses (visual recommendation) dresses in a similar color palette (visual recommendation) dresses with a similar cut and silhouette (visual recommendation) high waisted dresses with the same fit and fabric (alternate product recommendation) similar dresses with additional features like pockets (alternate product recommendation) personalized recommendations based on customer behavior, frequently bought items, and purchase history (get recommended items) this enhances product discoverability and customer satisfaction where recommendations appear product detail pages (pdp) – displays relevant product suggestions out of stock items – provides recommendations when selected items are unavailable more like this section – allows users to explore additional product recommendations personalized recommendation sections – tailored suggestions based on user interaction and feedback