This paper tackles the problem of minimising the direct server costs of complex multiple-server queueing systems while improving the quality perceived by customers. The problem is solved by finding the server roster, defined as the allocation of active individual servers in each time interval of the working day. As no closed solution is available at this time, it was obtained by simulation modelling. The complex queueing system was solved by a simulation-based heuristic that includes the dynamic arrivals, entity reneging and/or balking, multi-servicing, and the individual service-time distributions of servers. The distributions of the random variables of single or multiple-branch businesses were identified by sampling procedures for multiple company-branch deployment. The beneficial financial impact of the proposed solution is confirmed in numerical real-life banking scenarios. The quality perceived by customers (evaluated through the expected queueing time) was also improved. Finally, additional insights that might improve the system performance are highlighted.