Cameras capture time-stamped images of predefined areas. The predefined areas are assigned to one or more zones. The images are processed to recognize items and persons within the zones. Custom metrics with respect to each zone relative to at least time durations, items, and persons are logged in a data store. An interface is provided to custom define queries that are processed against the data store. In an embodiment, the data store is mined to discover relationships between the metrics for each zone, and real-time actions are processed when predefined conditions are discovered from the metrics.