Real-Time Bidding

Real-time bidding studies how to choose ad bids and allocate ad opportunities within milliseconds while meeting goals such as budget, return on ad spend, and platform revenue. Common approaches include pacing and dual-price control, model-predictive control, reinforcement learning, and generative or retrieval-based policies; models also account for first-price auction uncertainty and competition across channels. Work increasingly considers platform-side decisions such as ad allocation and request dispatch, alongside advertiser bidding. A central challenge is showing that offline replay results transfer safely to live auctions, where limited data support, changing competition, latency, and privacy-related information gaps can alter outcomes.