Automated Returns Analysis for a Multi $B Home Improvement Retailer

Client had enormous volumes of product returns data to manually categorize and interpret. This was a slow process that negatively impacted customer satisfaction and led to a lack of visibility to product development and quality control functions.

Our approach included first collecting a large dataset of returns and manually labeling a subset with the correct return classifications. Then, we applied natural language processing and supervised learning techniques to process and categorize the unstructured return descriptions into structured return categories. With a system in place to automatically classify consumer responses, the client had a data-empowered team. They could more quickly identify return trends, pull problematic or low-quality product from shelf and easily close the “learning loop” by sharing the insights with product development and marketing teams to make necessary changes.

The output of this model could then be redeployed in a predictive model to proactively identify buying trends and return trends. Teams gained insight into unfavorable, high-risk and low-quality product or categories of products and could pave a path for fewer returns and happier customers.