Techniques for self-learning an item's weight range are provided. Weights for items during transactions with an enterprise are recorded and plotted in a data store. A graph is used to plot weights for the items over time. During a particular transaction, when an item is identified and its weight is captured, the graph is consulted to detect any deviance in the captured weight to the graph. When a deviance falls beyond a threshold real-time alerts are raised within the enterprise to manually inspect the item and recorded weight before a transaction with a customer concludes.