An efficient process is provided that exploits features in historical and current client location data to forecast client counts of different zones up to several hours ahead. These features may be obtained from correlations of client counts of multiple zones in a recent and long period of time. These features may also be combined using techniques that choose the best performing method for a particular dataset and a particular lookahead time. This process provides better forecast/prediction on the zone-based client count data, and is very useful in customer analytics which can now show the future predicted value. This can help the analytics customers to plan their operations based on the location analytics.