Workforce capacity planning with hierarchical skills, long-term training, and random resignations

Abstract

This paper addresses a multistage capacity planning problem for a hierarchically skilled workforce in a production environment. Recruits are hired with little or no experience and are trained over multiple periods to perform jobs that require increasing levels of skill. Training can take place either off-the-job, on-the-job or a combination thereof. The problem is complicated by random resignations that can lead to labor shortfalls that jeopardise continuous operations. The objective is to balance workforce costs with penalty costs associated with skill shortages. The problem is modelled as a Markov decision process for which several parameterised decision rules are proposed to find solutions. A large-scale neighbourhood search is developed to deal with ‘noisy’ cost function measurements. Experiments show that good parameter values can be found in less than four hours using real-world data. When training requires extensive supervision, the results indicate that the number of workers concurrently in training should be limited. They also show that a shorter, intense training period during which employees do not perform regular tasks is generally preferable to a longer training period where employees spend time both on and off the job. Finally, we demonstrate the value of worker flexibility when downgrading is applied.