STACKR GROUP IN-STORE
SALES OPPORTUNITY COUNTING
STACKR GROUP IN-STORE SALES OPPORTUNITY COUNTING
In retail, up to 72% of in-store traffic can be “noise” salespeople, delivery drivers, passers-by that distorts conversion rate calculations. Stackr Group needed to accurately identify real sales opportunities from CCTV footage, at scale, across 150+ stores. My Data Machine built the Human-in-the-Loop pipeline that made it operational in just a few weeks.
By deploying HITL operators to validate customer counts across thousands of frames every month classifying pedestrians, groups, children, occlusions, and superimposed silhouettes My Data Machine created the reliable ground truth needed to train Stackr’s computer vision models across 50+ store layouts and 4+ retail banners.
The context
Stackr Group’s retail tech platform helps store owners measure in-store conversion rates with precision. But raw CCTV traffic data is inherently noisy without accurate classification of who is actually a potential customer, sales opportunity counts are unreliable and actionable insights become impossible.
The challenge
Processing 0.5M+ CCTV images per month across 150+ stores and 50+ different layouts required a scalable annotation pipeline capable of handling complex visual scenarios occlusions, groups, children, overlapping silhouettes with the consistency needed to produce trustworthy training data for production AI models.
