AI Models for Understanding Customer Needs and Guiding Product Innovation

restricted_to_institution
Restricted (Penn State Only)
Author:
Yu, Wenlan
Graduate Program:
Business Administration
Degree:
Doctor of Philosophy
Document Type:
Dissertation
Date of Defense:
May 26, 2026
Committee Members:
Brent Ambrose, Program Head/Chair
Min Ding, Major Field Member
John Liechty, Minor Field Member
Arvind Rangaswamy, Chair & Dissertation Advisor
Ning Zhong, Major Field Member
Kenneth Huang, Outside Unit & Field Member
Keywords:
product management
new product development
customer needs understanding
creativity
graph neural networks
large language models
product management
new product development
customer needs understanding
creativity
graph neural networks
large language models
Abstract:
Firms routinely make product decisions that depend on understanding customer needs and product relationships. Product managers seek to identify unmet customer needs and generate new product concepts, while retailers and digital platforms seek to understand which products serve as substitutes, which products are frequently purchased together in customers’ shopping baskets, and how customers’ co-purchase behavior evolves over time. These decisions inform a wide range of downstream marketing activities, such as product development, assortment planning, recommendation systems, inventory management, and merchandising strategies. However, making such decisions has become increasingly challenging in contemporary markets, where customer needs, competitive dynamics, and product relationships evolve rapidly. At the same time, advances in artificial intelligence (AI) and the growing availability of large-scale behavioral and unstructured data, such as product reviews, have created new opportunities to address these challenges. Yet many AI applications in marketing remain primarily focused on prediction, rather than on uncovering interpretable market structures that can support strategic product decision-making. This dissertation develops theory-grounded AI frameworks that learn structured and interpretable representations of product markets from large-scale consumer data. The central premise is that strategic product decisions can be improved when AI systems are designed to recover marketing-relevant structures, such as customer associative networks, product analogies, and complementarity and substitutability relationships. Rather than treating AI as a black box prediction tool, this dissertation uses psychology and marketing theories to guide graph representation learning and to generate outputs that are interpretable, actionable, and useful for managerial decision making. The dissertation studies two product management problems. Essay 1 examines opportunity identification and product ideation. It develops a Dual Bipartite Graph Neural Network (DBGNN) that constructs a market-level associative representation of products, product features, and usage situations from product descriptions and user reviews. This representation is used to retrieve structured analogical contexts for LLM-based product ideation. In an application to 2,700 Google Play apps, the approach increases the diversity and originality of generated ideas while maintaining strong ratings on incremental usefulness and consumer use intention. Essay 2 examines product complementarity and substitutability. It develops a temporal graph decomposition framework that dynamically estimates SKU-level product relationships from large-scale transactional data. By distinguishing products that are purchased together from products that compete with one another, the framework can support more adaptive decisions in assortment planning, recommendation design, cross-selling, and inventory management. Together, the two essays show how theory-grounded graph representation learning can support strategic product decisions by making market knowledge more structured, interpretable, and actionable. This dissertation makes three contributions. Conceptually, it contributes to research on theory-grounded AI in marketing by showing how existing theories and constructs can guide the design of AI systems that support strategic reasoning rather than focusing on prediction alone. Methodologically, it demonstrates how graph representation learning can recover interpretable market structures from large-scale consumer behavioral and textual data. Substantively, it develops AI-based approaches to mass product ideation and dynamic product relationship measurement, two product management challenges that have become increasingly important in data-rich and rapidly evolving markets.