Adaptive Supergeo Design: A Scalable Framework for Geographic Marketing Experiments
Adaptive Supergeo Design: A Scalable Framework for Geographic Marketing Experiments
Geographic experiments are a gold-standard for measuring incremental return on ad spend (iROAS) at scale, yet their design is challenging: the unit count is small, heterogeneity is large, and the optimal Supergeo partitioning problem is NP-hard. We introduce Adaptive Supergeo Design (ASD), a two-stage framework that renders Supergeo designs practical for thousands of markets. A bespoke graph-neural network first learns geo-embeddings and proposes a concise candidate set of 'supergeos'; a CP-SAT solver then selects a partition that balances both baseline outcomes and pre-treatment covariates believed to modify the treatment effect. We prove that ASD's objective value is within (1+epsilon) of the global optimum under mild community-structure assumptions. In simulations with up to 1,000 Designated Market Areas ASD completes in minutes on standard hardware, retains every media dollar, and cuts iROAS bias substantively relative to existing methods. ASD therefore turns geo-lift testing into a routine, scalable component of media planning while preserving statistical rigour.
Charles Shaw
地理计算技术、计算机技术
Charles Shaw.Adaptive Supergeo Design: A Scalable Framework for Geographic Marketing Experiments[EB/OL].(2025-06-30)[2025-07-20].https://arxiv.org/abs/2506.20499.点此复制
评论