Companies often struggle to fix underperforming products, even when they have an abundance of consumer product evaluations in the form of reviews. New research from the University of Florida Warrington College of Business introduces an AI-driven approach that transforms fragmented and ambiguous customer feedback into precise, actionable product improvements, pinpointing the minimum set of feature changes needed to boost market performance.
Tested on real-world reviews of frying pans on a prominent U.S. retailer’s website, the method achieved 73% precision, outperforming existing approaches by about 11%, and produced product recommendations that consumers consistently preferred over original designs.
“Customer reviews are a key source of evaluation for product managers and consumer insights teams,” explained Amin Hosseininasab, Assistant Professor of Marketing at UF Warrington. “The challenge is not a lack of data about these products, but how to use the data effectively.”
Companies looking to improve unsuccessful products face several challenges when searching for opportunities for improvement. These products are purchased less often, resulting in limited data. When feedback is available, it is often vague (such as “poor quality”) without identifying which specific features should change. Consumer perceptions can also be shaped by how features interact, making it difficult to isolate root causes. At the same time, not every issue needs to be addressed, as products can often be improved by changing only a few attributes.
To address these challenges, paper authors Hosseininasab, Matherly Professor of Information Systems Anuj Kumar, and Ph.D. student Vincent Zhao, developed a Product Segmentation Tree algorithm that identifies which shared product features are responsible for poor product performance across the market. Grouping products based on shared features and performance, the algorithm identifies which features lead to high or low product performance. By developing an additional algorithm, the research team can also identify the shortest path (through the minimum number of feature changes) from unsuccessful products to a successful product.
“The product development team can use this algorithm to identify which product features to change, and to which specific values, to improve the product’s market performance,” said Hosseininasab. “The PS Forest algorithm’s performance on real-world data is a testament to its ability to address key challenges in leveraging customer reviews to provide interpretable and actionable decision support for product improvements.”
