Part G Marketing Analytics , Marketing Metrics , and Research Track Chairs

Research Question As a clothing and accessories retailer, Zara launches 11,000 designs on store shelves every year. There are over 200 distinct items in a given Zara store with a one-week inventory turnover. A retailing chain like Zara usually maintains a wide variety of products at its warehouse, but only displays a much smaller set of products at a given store due to space constraints. Even so, evaluating all the items (>200) in a single store would be a tremendous drain on time and require significant cognitive effort from customers. Since only a subset of the available inventory is typically evaluated, a consumer may leave the store without finding anything s/he likes, even when the retailer may possibly carry the products appealing to his or her preferences—at the store, a different store, or the warehouse. Understanding customer preference is the key to address these problems in garment retail stores. We have developed and tested a model relying on the state-of-art computer vision techniques, to infer customers' individual preferences based on their affective and behavioral responses, which are captured by video recording. The method infers a customer preference for garment features from physical clues such as facial expressions and features-of-interest on garments. In this paper, we propose and test one method of inferring individual preferences using in-store video analysis, relying on the state-of-art computer vision techniques. This video-based preference inference method automatically detects customer's facial expressions and features-of-interest from video captured while a customer is evaluating garment in a retail store. Customers' preferences on a subset of garment features are then inferred from this information. These partial preferences, together with customer's body features extracted from the video are then used to match with a customer database to construct a composite preference profile for the focal customer. Finally, product recommendations are made for customer based on this composite preference profile. The recommendations help customers find products that fit their preferences and thus enhance their satisfaction of overall shopping experience, thereby increasing store sales in both the short and long run. This model combines video analysis method from the computer science discipline and standard marketing research methods. We test performance of the garment recommendation model in a garment retailing context, where 127 apparel shoppers choose from 140 garments displayed on racks in a store-like environment. The model is able to make individualized recommendations every time a customer evaluates a product, and its superior …