The paper deals with experimental characterization and analysis of an isolated delivery vehicle fleet system of a retail company. The on-vehicle data collection has been conducted for a fleet of ten delivery vehicles running continuously over a three month period. Next, a wide statistical analysis of the collected data is presented, in order to provide a basis for future investigation of possible benefits of replacing the conventional vehicle fleet with a hypothetical one based on electric vehicles. Finally, the recorded large set of driving cycles is used for the purpose of stochastic-based synthesis and validation of a small number of representative driving cycles. Such naturalistic driving cycles can further be used for electric vehicle configuration optimization and controller design, as well as for design and verification of a vehicle fleet model aimed for various energy planning and smart charging studies.