A Markov Decision Model for Managing Display-Advertising Campaigns.

This article focuses on developing an analytical methodology for ad agencies to optimize bidding and viewer-allocation policies in digital display advertising campaigns, aiming to maximize agency profits while meeting campaign goals under uncertainty. The problem is modeled as a Markov decision process (MDP) that accounts for stochastic arrivals of viewers and ad campaigns and uncertain bid outcomes on ad exchanges. The authors derive exact dynamic, state-dependent solutions for both finite horizon and steady-state cases, and propose a heuristic method that simplifies computation with minimal profit loss, making it practical for large-scale problems. Numerical analysis based on empirical data validates model assumptions and demonstrates that dynamic bidding policies outperform static ones, while controlling the capacity of impression queues emerges as an effective managerial lever to balance delay penalties and bidding costs. The study also highlights operational benefits from consolidating campaigns and optimizing queue capacities, providing insights relevant for campaign managers and digital advertising platforms.