
AWS provides guidance for intelligent product substitution recommendations using OpenSearch Service and machine learning embeddings to help retailers offer similar alternatives when items are out of stock, reducing lost sales and improving customer experience.


<div><p>This article presents AWS guidance for intelligent product substitution recommendations to address out-of-stock retail items using machine learning and vector embeddings.</p><ul><li>Grocers face 8.2% average out-of-stock rates, risking $7-12B in sales annually</li><li>Uses Amazon OpenSearch Service k-NN index with text embeddings for product similarity matching</li><li>Employs all-MiniLM-L6-v2 sentence transformer to convert product titles/descriptions to numerical vectors</li><li>Architecture: S3 upload → Lambda → DynamoDB → Lambda embedding → OpenSearch indexing</li><li>API Gateway endpoint enables substitution queries with optional filters (category, price, allergens, diet type)</li><li>Supports JSON Lines product catalog format with required id and title fields</li><li>Recommends showing top three substitutions to order pickers for selection</li></ul><p>This solution helps retailers reduce lost sales from stockouts by providing intelligent, filtered product recommendations based on semantic similarity.</p></div>


Related articles