More Ads, Happier Shoppers: Unified-Valuation Ad Allocation at Scale
Paper |
Sergey Kolbin, Arushi Jain, Manavender Malgireddy, Kurt Zimmer, Masoum Mosmer, Patrick Phelps
Balancing short-term ad revenue with long-term user satisfaction is a central challenge in sponsored product search. We propose common currency, a unified ranking framework that merges sponsored and organic results within the same valuation. By translating predicted clicks and conversions into variable contribution dollars (VCD), both ad and organic items compete on a single, profit-based metric. We combine this objective with guardrails on relevance and ad load to preserve a high- quality customer experience. Our approach is implemented via a one-pass, order-preserving greedy merge, which we show is near-optimal in offline simulations despite the problem’s inherent non-monotonicities. We also compare real-time and batch variants, finding that batch ranking recovers most of the profit uplift while incurring significantly less latency. In a six-week online experiment covering millions of keyword searches, our method allowed to expand ads footprint and increased ad revenue by 35% and, surprisingly, also improved conversion rates and long-term profit across ads and organic products together. These results suggest that when ads are relevant and well- spaced, elevating ad load can benefit both advertisers and end users, challenging the classic trade-off narrative. We discuss lessons learned from production deployment and outline directions for further enhancements.