Walmart's InvAwr-RAG boosts ad fill rate by 68% with inventory-aware AI

Sponsored search is a pillar of e-commerce revenue, but many user queries fail to match any advertiser keywords, leaving ad slots empty and revenue on the table. To close this gap, Walmart researchers (Md Omar Faruk Rokon, Weizhi Du, Zhaodong Wang, Musen Wen) developed InvAwr-RAG (Inventory-Aware RAG), a generative AI model that combines retrieval-augmented generation with real-time inventory data. The model dynamically crafts new queries or rewrites existing ones to align with available ad campaigns, solving the huge variance in query intent and the massive, evolving keyword landscape.

In preliminary tests on Walmart's platform, InvAwr-RAG achieved a 68% increase in ad fill rate—the proportion of queries that serve at least one ad—while maintaining balanced relevance metrics. The model blends historically successful queries with novel, inventory-sensitive generations, ensuring that rewritten queries are both relevant to the user and commercially viable. Published at eCom@SIGIR 2024, this approach sets a new standard for dynamic query optimization, promising significant gains in advertiser ROI and user experience while turning previously unmonetizable searches into revenue opportunities.