Dynamic Segmentation for Large-Scale Marketing Optimization

Marketing organizations often design multiple marketing campaigns to meet a variety of objectives. Marketing resources must be allocated carefully across campaigns to minimize conflicts and maximize organizational value (e.g., impact on customer lifetime value). While data-driven analytics has greatly improved the ability to predict the effect of marketing actions on individual customers, such fine-grained discrimination of customers has made multi-campaign optimization much more difficult. We propose a new dynamic segmentation approach for linear programming in largescale, multi-campaign targeting optimization. Using on a novel form of column generation, our method produces segments of customers who are provably treated identically in the optimal solution of the underlying (unsegmented) problem. The resulting compression allows problems involving hundreds of campaigns and millions of customers to be solved optimally in fractions of a second. We also describe how the data-intensive components of the algorithm can be distributed to take advantage of modern clustercomputing frameworks.