Market Segmentation Trees

Problem definition: We seek to provide an interpretable framework for segmenting users in a population personalized decision making. Methodology/results: propose general methodology, market segmentation trees (MSTs), learning segmentations explicitly driven by identifying differences user response patterns. To demonstrate the versatility of our we design two new specialized MST algorithms: (i) choice model (CMTs), which can be used predict user’s amongst multiple options, and (ii) isotonic regression (IRTs), solve bid landscape forecasting problem. theoretical analysis asymptotic running times algorithmic methods, validates their computational tractability on large data sets. also customizable, open-source code base training MSTs Python that uses several strategies scalability, including parallel processing warm starts. Finally, assess practical performance synthetic real-world sets, showing method reliably finds accurately behavior. Managerial implications: The standard approach conduct making is first perform clustering according similarities contextual features then fit “response model” each segment how respond decisions. However, this may not ideal if prominent distinguishing clusters are key drivers Our addresses issue integrating modeling, consistently leads improvements prediction accuracy, thereby aiding personalization. find such integrated computationally tractable effective even large-scale Moreover, because segments easily described tree often require only fraction number generated traditional approaches. Disclaimer: This work was done prior Ryan McNellis joining Amazon. Funding: supported National Science Foundation [Grants CMMI-1763000 CMMI-1944428]. Supplemental Material: online appendices available at https://doi.org/10.1287/msom.2023.1195 .