| dc.description.abstract | Marketing researchers and practitioners care about why and how products or services are adopted by consumers. The influential theory of innovation diffusion has been established for decades, but modeling and simulating the diffusion process remains notoriously challenging. Lately, agent-based models (ABMs) have dominated traditional aggregate diffusion models, due to the remarkable advantage to capture individual heterogeneity and social and spatial interactions. Our critical review of the empirically-grounded ABMs of innovation diffusion, however, reveals that few such ABMs are calibrated properly, validated rigorously, and developed explicitly for prediction. This clearly limits their use in supporting decision-making in practice. The thesis contributes a rigorous data-driven agent-based modeling (DDABM) approach that relies on state-of-the-art machine learning techniques to effectively calibrate and validate agent behavior models on massive and rich individual adoption data. The models are integrated into multi-agent simulations to precisely forecast roof-top solar adoption and efficiently explore subsidizing strategies in San Diego county, USA. Historically, ABMs were used to answer “what-if” questions and draw insights on the efficacy of different policies, however, few could provide executable and quantitative decisions. Mathematical optimization has been widely used to provide numerical solutions in many domains, but little effort has been made to couple it with ABMs. By solving marketing optimization problems in several important settings, such as, dynamic seeding of emerging technologies, route planning for door-to-door targeted marketing, and budget optimization in multi-channel marketing, the thesis also strongly demonstrates how efficient algorithms can aid the design of effective marketing policies facilitated by data-driven simulations, like ABMs, providing optimal or near-optimal actionable plans for marketers. The presented research characterized by computational modeling techniques and algorithmic methods could lead to our ultimate goal of intelligent machine-automated marketing. | |
