Over the past decade, marketing environments have become increasingly interconnected, generating complex relational structures among firms, consumers, and market intermediaries. These interconnected relationships often produce spillover effects, coordination dynamics, and collective outcomes that cannot be fully understood through traditional unit-level analysis. Recognizing this limitation, this dissertation adopts a network-based perspective to examine how interactions among market participants shape marketing performance and organizational outcomes. Across two essays, this dissertation develops novel analytical frameworks that combine econometric reasoning with advanced deep learning techniques to model marketing systems characterized by rich relational dependencies. The first essay examines collaboration and knowledge sharing in team-based sales environments. In many modern organizations, particularly in industries such as insurance and financial services, sales agents frequently form temporary and overlapping teams to sell complex products. These collaborative structures create rich interdependencies among agents, where team formation, within-team collaboration, and cross-team knowledge spillovers jointly influence individual and team performance. To capture these dynamics, this essay introduces a novel Two-Stage Hypergraph Neural Network (TSHyGNN) framework that models agents as nodes and collaborative teams as hyperedges. Unlike traditional dyadic network models, hypergraphs allow multiple agents to simultaneously belong to the same collaborative structure, thereby capturing the multidimensional nature of real-world team interactions. A key methodological challenge in studying team collaboration is the endogeneity arising from agents’ strategic decisions regarding team formation. To address this issue, the proposed framework incorporates a two-stage estimation strategy that embeds a correction term within the performance prediction stage. The model also introduces a convolution architecture that integrates external network structures as holistic covariates, enabling richer representations of agents’ collaborative environments. Empirical results demonstrate that the proposed framework significantly outperforms conventional predictive models. Beyond predictive performance, the model enables counterfactual simulations that provide actionable managerial insights regarding team design, collaboration strategies, and resource allocation within sales organizations. By integrating econometric thinking with state-of-the-art hypergraph deep learning, this essay contributes both methodological innovations and substantive insights to the marketing and salesforce management literature. The second essay shifts the focus to spatial interactions within retail environments, examining how physical layouts influence consumer traffic patterns and store performance in shopping malls. Shopping malls represent a classic two-sided market in which mall owners seek to optimize overall traffic flow and tenant performance, while individual retailers aim to maximize their own store-level outcomes. Designing effective mall layouts therefore requires understanding how spatial positioning and consumer movement patterns generate network effects across retail units. Despite the importance of spatial interactions, most existing research treats stores as independent units and overlooks the networked nature of consumer movement within retail spaces. To address this gap, the essay constructs a comprehensive network representation of a shopping mall using large-scale foot traffic data collected through AI-enabled cameras. In this network, retail stores and service facilities such as elevators and restrooms are modeled as heterogeneous nodes, while consumer movement between locations forms directional edges capturing traffic flows. Building on this representation, the study develops a Graph Neural Network (GNN) framework that jointly predicts store-level sales and directional traffic flows through multiple loss functions. By incorporating spatial layout information as structural covariates, the model captures how mall design influences both consumer movement and retail performance. The framework also enables counterfactual simulations of alternative layout configurations, such as relocating service facilities or modifying store category clustering. These simulations generate actionable insights for mall operators seeking to optimize tenant performance while maintaining balanced traffic distribution across the retail environment. Together, the two essays highlight the importance of adopting a network perspective in marketing research. Marketing systems frequently involve complex relational structures in which the actions of one agent influence the outcomes of many others. Whether through collaborative sales teams or spatial interactions among retail locations, these interconnected dynamics generate spillovers that cannot be adequately captured by traditional models that treat observations as independent. By developing AI-enhanced modeling frameworks that integrate econometric methods with graph-based deep learning, this dissertation advances our ability to analyze such relational marketing systems.