Embodiments of the present invention provide techniques for identifying and quantifying waste in a process. Waste information is input via images and/or natural language. The amount of waste is estimated based on information in images and/or a natural language description. A computer-implemented technique extracts metadata on waste products from the images and/or natural language description. A variety of factors such as social media trends, weather, traffic, and/or sports schedules are evaluated by the computer and used in predicting the amount of waste that will occur.