טכניון מכון טכנולוגי לישראל
הטכניון מכון טכנולוגי לישראל - בית הספר ללימודי מוסמכים  
M.Sc Thesis
M.Sc StudentArik Senderovich
SubjectMulti-Level Workforce Planning in Call Centers
DepartmentDepartment of Industrial Engineering and Management
Supervisor Full Professor Gal Avigdor


Abstract

In large service organizations, such as call centers or hospitals, workforce is planned mostly at two levels: top-level planning that corresponds to long-term planning periods (months, quarters, years) and low-level planning that corresponds to short-term day-to-day shift-staffing.


In top-level planning, the number of agents to be employed accounts for various workforce aspects such as: hiring, skill-levels, promotions and turnover. Low-level detailed planning is concerned with staffing agents to upcoming shifts so that short-term demand for service will be adequately met. We argue that in an organization with highly varying demand for service (such as a Call Center), it is impossible to separate the two planning horizons. Therefore, we propose to follow a multi-level hierarchical framework that will account for both top-level and low-level planning.


The main goal of our research is thus to develop a methodology for multi-level workforce planning and then validate our theory in a real Call Center. In order to pursue our goal, we adopt a multi-level workforce planning model. We first apply a relaxed version of the DP model, to serve as a benchmark for more complex models. We then apply the full DP model and propose an extension that better describes our Test-Case Call Center.


Model parameters are estimated by applying statistical techniques to real daily-updated Call Center data, provided by the Technion SEELab. Parameters related to agent utilization profiles are inferred via data-driven server networks. These networks describe agent activities at the transaction-level resolution, thus eliminating the need for classical IE work sampling and time-and-motion techniques.


A generalization of the data-driven server networks that we referred to as service networks is also discussed. These networks are defined as data-driven realizations of the service process. Various applications to service networks are presented.


Lastly, we discuss aspects of model validity, by comparing our results to the workforce planning in our Test-Case Call Center. We provide evidence in favor of the multi-level framework and explain the need for model-updates in a rolling horizon fashion.