Retail Intelligence in Action (RetInA) is an IoT solution designed to perform real-time store analytics.
‘Retailers are not capturing consumer behavior’ – Forrester, 2016 Against this backdrop, Microsoft Retail Store division wanted a solution that would deliver real-time, actionable insights into customer engagement and assist retailers in transforming the store experience.
RetInA was interesting to build because of the challenges in quantifying customer engagement in physical environment vis-à-vis online – where each product-page activity is logged. In physical space, data cannot be collected for every customer without explicit opt-in, thus precluding existing enablers (loyalty programs, mobile web tools) that forgo huge amounts of data during large visitor flux.
I focused on three key aspects of the solution – store-floor division into product-zones to obtain customer trajectory in each; tracking individual customer-coordinates though sensors; aggregating and translating data into rich visualizations that provide actionable insights into customer’s interest.
The customers were unobtrusively tracked as blobs using depth and grayscale feed from an IoT motion sensor, hence pre-empting gathering of any personally identifiable information. The spatial (x,y) coordinates of each customer on store-floor was extracted using computer-vision techniques like contour-mapping and homography, and by differential processing of captured frames. This approach resulted in a compact space-time data stream that enabled faster transfer to cloud for real-time experience. Furthermore, Kalman Filter was employed to resolve ambiguity in scenarios where customers overlap/cross each other, bringing the individual customer tracking accuracy to ~87%. The insights were reported on cloud via interactive visualization tools of PowerBI – Microsoft’s cloud analytics engine – making the insights available anywhere, anytime.
RetInA was well-received during initial trials, with plans for further installations.