For most physical stores, procuring customer behavior data and client management is difficult yet crucial. For stores, especially luxury stores, who value returning customers the most, identifying customer’s identity and past purchase history would significantly improve the sales by personalizing the recommendation. Many chain stores utilize membership card to collect purchase information and change sales strategy accordingly, yet overlooking the vast majority of customers who are not registered members. More detailed statistics including age, gender, customer counts are extremely valuable for adjusting business strategy, which brings the importance of a real-time customer analytic system for retail business. We have developed an accurate, real-time and fully automated system that relies on computer vision and deep learning to identify returning customers and comprehensively filter sales and customer information. To adjust to network bottleneck, limited computation resources, low latency requirement, recognition accuracy and other difficulties, we divided the system into two submodules. The system depends on local computation resources to perform customer detection, tracking and feature extraction. Whereas age and gender estimation, and customer recognition are performed on cloud computing resources to compensate for the limited computation resource at local store and to accommodate multiple stores requirements.