Recent developments in energy metering technologies have allowed electric load data to be more easily accessible. Services that use this data to inform customers can raise awareness about electricity consumption and provide suggestions to encourage energy efficient behavior. Quantifying the flexibility of the demand combined with accurate predictions of the total electric load allow information services to provide suggestions to end-users on how to reduce electric consumption that are appliance and time specific. With the arrival of new electric generation technologies, such as photovoltaic or wind energy, demand side flexibility will play an important role in the optimization of the future electric system. The accurate prediction of demand flexibility at a building level, therefore can contribute to the optimization of load scheduling. This study presents an effective multi-step technique to forecast the average hourly demand flexibility for a household, using neural networks, unsupervised clustering techniques and an optimization algorithm, combined with statistical studies. The model is mainly based on the historical electric use of a building and does not require significant computational capacity, thus making it widely applicable and informative for residential customers, helping them improve their behavior to be more energy efficient in the future.