HMAF: A Hierarchical Multi-Slot GD-RTB Allocation Framework

Researchers propose HMAF, a hierarchical allocation framework that optimizes ad impression distribution across guaranteed delivery contracts and real-time bidding auctions. Implemented at Meituan, the system achieved a 3.72% improvement in delivery rates and 1.59% revenue increase by unifying previously decoupled optimization approaches.

HMAF addresses a fundamental operational challenge in digital advertising platforms: balancing contractual obligations with dynamic revenue opportunities. Traditional systems treat guaranteed delivery (GD) contracts and real-time bidding (RTB) auctions as separate problems, leading to suboptimal resource allocation and missed revenue opportunities. This framework integrates both through a three-stage approach: offline planning establishes GD resource constraints, dynamic calibration adjusts competitive positioning between GD and RTB, and online execution makes real-time allocation decisions across multiple ad slots.

The research reflects broader industry trends toward unified optimization in ad tech. As platforms mature, siloed systems become increasingly inefficient—GD contracts ensure predictable revenue but constrain flexibility, while RTB maximizes short-term gains but risks contract breaches. Meituan's scale as a food delivery platform with millions of daily impressions makes this optimization particularly valuable, as small efficiency gains compound significantly.

The reported outcomes—3.72% delivery rate improvement and 1.59% revenue growth—indicate meaningful commercial impact. These metrics suggest the framework reduces both contract violations (costly for platform reputation and client relationships) and opportunity costs from inefficient allocation. For ad tech platforms globally, this demonstrates that sophisticated allocation algorithms can simultaneously improve customer satisfaction and profitability.

Future applications likely extend beyond food delivery to e-commerce, search, and social platforms facing similar GD-RTB dynamics. The framework's emphasis on multi-slot environments addresses real-world complexity where impressions vary in quality and targeting capability. As machine learning becomes central to ad operations, unified optimization approaches will differentiate platform competitiveness.

• →HMAF integrates guaranteed delivery and real-time bidding optimization through offline planning, dynamic calibration, and online execution.
• →Implementation at Meituan achieved 3.72% improvement in contract delivery rates and 1.59% total revenue growth.
• →The framework addresses the industry challenge of balancing contractual obligations with dynamic revenue maximization.
• →Multi-slot allocation and complex impression constraints require unified rather than siloed optimization approaches.
• →Results suggest sophisticated ad tech allocation algorithms can simultaneously reduce contract violations and maximize platform revenue.