Meituan's HMAF Framework Boosts GD Ad Delivery by 3.72%
New hierarchical framework balances guaranteed delivery and real-time bidding for 1.59% revenue lift
Online ad platforms often juggle Guaranteed Delivery (GD) contracts—where advertisers pay for a fixed number of impressions—alongside Real-Time Bidding (RTB) auctions. Prior approaches either optimized GD and RTB separately or relied on heuristic priority rules, leading to suboptimal trade-offs between short-term revenue and long-term contract fulfillment, especially in multi-slot environments with impression constraints.
HMAF (Hierarchical Multi-Slot Allocation Framework) tackles this with a three-stage Plan-Calibrate-Execute paradigm. It first plans GD resource allocation offline, then dynamically calibrates GD-RTB competitiveness, and finally makes real-time listwise rank decisions across multiple ad slots. Deployed at Meituan—one of the world's largest food delivery platforms—HMAF lifted GD delivery rate by 3.72% and total advertisement revenue by 1.59%. The work is accepted at KDD 2026 (Applied Data Science Track) and co-authored by researchers from Meituan and academic institutions.
- HMAF uses a Plan-Calibrate-Execute paradigm to integrate offline GD planning with online RTB decision-making
- Deployed at Meituan, it increased GD delivery rate by 3.72% and total ad revenue by 1.59%
- Accepted at KDD 2026 Applied Data Science Track; co-authored by researchers from Meituan and academia
Why It Matters
Balances guaranteed ad delivery with real-time bidding, boosting both fulfillment rates and revenue for large-scale ad platforms.