PlatformBid Benchmark Optimizes Auto-Bidding for Unified Ad Platforms

Traditional auto-bidding algorithms in computational advertising primarily focus on maximizing advertiser conversions within the Demand Side Platform (DSP) context. However, major advertising platforms today, such as social media and e-commerce giants, integrate the Supply Side Platform (SSP), DSP, and Ad Exchange functions internally. This unified structure means their auto-bidding goals extend beyond advertiser success to also include maximizing the platform's overall revenue. To address this evolving landscape, a new benchmark called PlatformBid has been developed. It provides a comprehensive framework for evaluating auto-bidding algorithms from a unified ad platform's perspective. PlatformBid defines three realistic competitive settings: homogeneous competition (identical algorithms), heterogeneous competition (diverse strategies), and promotional competition (budget surges during events like Black Friday). The benchmark systematically evaluates various auto-bidding methods, including classical control, reinforcement learning, and generative approaches. Additionally, a novel flow-matching-based method called BidFlow is proposed, demonstrating improved target cost in online experiments. PlatformBid aims to advance auto-bidding research by providing a more holistic and realistic evaluation environment.