Predicting Consumer In-Store Purchase Using Real-Time Retail Video Analytics

The proliferation of cameras and their video data in retail marketing presents new opportunities for academics to study customer behavior with the newer video analytics tools. In collaboration with a large retail chain store in Asia, we obtained a unique video dataset collected from in-store cameras and combined it with customer-transaction data. By leveraging state-of-the-art computer vision techniques, we extracted features of customer demographics, physiological appearance, emotional expression, and contextual dimensions from the videos. We implemented facial-recognition and face-tracking algorithms to extract consumer behavior with a limited amount of human aid and obtained consumer facial features on a scalable basis. We propose herein a novel framework that can use machine learning and deep learning models to analyze combined video and customer-transaction data in any commercial context to predict customer purchase decisions. The results show that our framework could in fact be effectively used to make predictions of consumer offline purchase decisions, which successful outcome reveals the importance of incorporating emotional response into prediction. Overall, our study demonstrates how video-based content can be used to understand customer behavior along multiple dimensions on a scalable basis. Our findings 1) complement the literature that examined customer behavior by incorporating video data into analysis, 2) reveal the multi-dimensional drivers of purchase decisions in a retail setting, and 3) provide for an implementable video analytics tool that can be usefully employed by marketers and practitioners. An important managerial implication, furthermore, is that our framework can be incorporated into the omni-channel retailing context to provide a win-win for both firms and customers and generate possibilities for offline recommendations