This paper develops a data-driven model for assessing the availability of flexibility from individual household devices, at house level. The model predicts the potential shift, increase or decrease of the energy consumption of a particular type of device at a given time, in response to a price signal, and for each house separately. Therefore, the location of the flexibility source is known with accuracy. The model has an Auto-Encoder architecture based on Convolution Neural Network. The augmented model demonstrates good performance in terms of predicting the time-shift in load.