[Paper Review] AHBid: An Adaptable Hierarchical Bidding Framework for Cross-Channel Advertising

Xinxin Yang, Yangyang Tang+4 more|arXiv (Cornell University)|Feb 26, 2026

Consumer Market Behavior and Pricing0 citations

TL;DR

AHBid combines a diffusion-model-based high-level planner with a control-based low-level bidder to optimize cross-channel auto-bidding under budgets and CPC constraints, achieving notable ROI gains over baselines. Online tests show about 13.57% improvement in overall return.

ABSTRACT

In online advertising, the inherent complexity and dynamic nature of advertising environments necessitate the use of auto-bidding services to assist advertisers in bid optimization. This complexity is further compounded in multi-channel scenarios, where effective allocation of budgets and constraints across channels with distinct behavioral patterns becomes critical for optimizing return on investment. Current approaches predominantly rely on either optimization-based strategies or reinforcement learning techniques. However, optimization-based methods lack flexibility in adapting to dynamic market conditions, while reinforcement learning approaches often struggle to capture essential historical dependencies and observational patterns within the constraints of Markov Decision Process frameworks. To address these limitations, we propose AHBid, an Adaptable Hierarchical Bidding framework that integrates generative planning with real-time control. The framework employs a high-level generative planner based on diffusion models to dynamically allocate budgets and constraints by effectively capturing historical context and temporal patterns. We introduce a constraint enforcement mechanism to ensure compliance with specified constraints, along with a trajectory refinement mechanism that enhances adaptability to environmental changes through the utilization of historical data. The system further incorporates a control-based bidding algorithm that synergistically combines historical knowledge with real-time information, significantly improving both adaptability and operational efficacy. Extensive experiments conducted on large-scale offline datasets and through online A/B tests demonstrate the effectiveness of AHBid, yielding a 13.57% increase in overall return compared to existing baselines.

Motivation & Objective

• Motivate cross-channel auto-bidding under budget and CPC constraints in dynamic advertising environments.
• Develop a two-tier framework that leverages generative planning and real-time control to improve adaptability and ROI.
• Capture temporal patterns and historical dependencies to stabilize decision-making and optimize allocations across channels.
• Demonstrate effectiveness through large-scale offline datasets and online A/B testing.

Proposed method

• Introduce a cross-channel constrained bidding (c3-bidding) problem formulation with J channels and stage-wise goals.
• Propose AHBid with a high-level planner (diffusion-model based) generating goal trajectories across M stages per channel.
• Incorporate a constraint enforcement mechanism to adaptively penalize constraint violations and a trajectory refinement (inpainting) using history to improve responsiveness.
• Develop a control-based bidder combining a historical MPC-inspired dual-variable model and a real-time LP refinement to adjust bidding under CPC and budget constraints.
• Train the planner via conditional diffusion modeling with classifier-free guidance and a hybrid loss that blends diffusion objective with constraint-violation signaling.
• Enable bidder adaptation by learning dual-parameter updates from historical data and refining them with online information; combine historical and real-time outputs with a weighted scheme.

Figure 1 . Illustration of auto-bidding system applications in multi-channel scenarios.

Experimental results

Research questions

• RQ1 Can a diffusion-model-based planner effectively capture temporal patterns and historical context to allocate budgets and constraints across cross-channel channels?
• RQ2 How can constraint satisfaction be ensured and improved when using generative planning in online bidding?
• RQ3 Does an adaptive, two-model bidding approach (historical + real-time) improve responsiveness and ROI in dynamic cross-channel advertising environments?
• RQ4 What is the empirical performance of AHBid compared to state-of-the-art baselines on both offline datasets and online A/B tests?

Key findings

• AHBid yields a 13.57% increase in overall return over baselines in online experiments.
• In offline evaluations, AHBid and its variants outperform PID, USCB, CQL-based MCQ, ABPlanner, DiffBid, and HiBid across AuctionNet and GenB-4c datasets.
• AHBid achieves higher R (returns) and improved CSR (constraint satisfaction rate) and BCR (budget consumption rate) metrics relative to competing methods.
• The diffusion-based planner effectively generates stage-wise goals that lead to more stable and higher-value allocations across channels.
• The combined historical+real-time bidding controller improves adaptability to environmental changes, sustaining performance under dynamic CPC and budget conditions.

(a) Comparison of Returns