Resolving Price-Segment Blindspot in Search Ranking for Value-Driven Marketplaces

Abstract

On value-driven marketplaces, user-product interactions concentrate heavily around affordable products. Search ranking systems trained on such engagement data therefore learn relevance patterns that primarily reflect the preferences of the dominant value conscious user cohort, limiting their ability to accurately personalize results for users whose intent aligns with relatively higher-priced products. We refer to this failure mode as the Price-Segment Blindspot. This challenge cannot be addressed through static promotion of higher-priced products, because price preference is highly contextual. A value-conscious user may occasionally search for an expensive ''silk saree'' before a wedding, while users with affinity toward high-priced products may still prefer inexpensive products for routine purchases. Effective ranking therefore requires modeling latent user affinity and real-time query intent to infer when different price segments are relevant. We address this challenge by incorporating user-level affinity and query-level intent signals into a production ranking system. However, on-platform interactions alone provide weak evidence of user's price-segment affinity due to sparse interactions in higher price segment. To address this sparsity, we augment the user affinity model with off-platform behavioral signals that better capture user spending patterns. In a three-week A/B experiment on the Meesho platform with ~300 million monthly active users, our approach improved the GMV_c (Gross Merchandise Value Contribution) of the High Price-Segment products by 5.92% and orders by 6.32%, while keeping the overall O/V_i (Orders-per-Viewer) nearly flat +0.06%.