An enormous amount of operational data can in theory be collected (measurement) at the (interactive) voice response unit (VRU) and the automatic call distributor (ACD). In practice, only summary statistics of such data are often stored, for example at the granularity of 30 minutes, in order to reduce the necessary storage space. The summary statistics usually include the number of arrivals and abandoned jobs, average service time, agents' utilization, and the distribution (percentiles) of waiting time in the queue .
The collected data are used to estimate the arrival process, service time distribution, and distribution of the time a customer is willing to wait, for each ``type'' of calls . The history of these estimates are then used to forecast future arrivals, service times, and the time willing to wait [30,118].
Forecasting allows capacity planning via a queueing analysis. Assuming that a routing policy at the ACD is given, a contact center can be modeled as a (multiserver) queueing system, where the ACD has queues of calls that are to be served by servers (agents). An analysis of the queueing system allows us to determine the number of agents (of each type and level of skill) to be assigned for each time slot (e.g. half an hour), and it also provides lessons and guidelines that are useful in designing a routing policy at the ACD. Simulation may/should be used for fine tuning the number of agents to be assigned. In Sections 8.3.2-8.3.3, we will elaborate on capacity planning and routing policy design.
Once the number of agents to be assigned for each time slot is given, an assignment of individual agents to the time slots can be determined. In determining the assignment, one usually needs to solve an integer program (IP), so that various constraints are satisfied. We study a similar problem in the context of nurse scheduling in . The solution to the IP provides the working schedule of each agent, and it can also be used for hiring and planning training. For example, if there are no feasible solutions (satisfying all the constraints), new agents need to be hired, or some existing agents need to be trained for new skills. The contact center operates following the assignment schedule, and the operational data will be measured, which in turn provides feedback to capacity planning and agent assignment.