Optimal forecasting of ATM cash repository in an optimal way is a complex task. This paper deals with cash demand forecasting of NN5 time series data using neural networks. NN5 reduced Dataset is a subsample of 11 time series of complete dataset of 111 daily time series drawn from homogeneous population of empirical cash demand time series. Main objective of this paper is to forecast cash demand forecasting of NN5 data with neural networks. Further, the same process is applied on clusters of ATMs. Discrete time wrapping is used as distance measure. Root mean square error has been calculated for such clustered group of ATMs and average is calculated. Root Mean Square error indicates applications of clustering before applying Neural Network increases precision in forecasting of ATM Cash Repository.