Automated Manpower Rostering: Techniques and Experience

Manpower is a highly critical resource in many industries. It is not surprising that organizations place substantial investment into manpower management matters. Manpower rostering, which addresses the issues of employing, allocating and scheduling manpower resources to meet operational demands, is at the heart of the manpower management cycle. Rostering is a particularly prominent activity in service organizations where employees need to be scheduled across several workshifts to handle varying customer demands 24 hr a day and seven days a week. The generation of rosters is a complicated process of meeting demands with available resources subject to a variety of work policy constraints and goals. The need for automated manpower rostering systems is thus intuitive and appealing. The system would reduce time spent on creating and modifying rosters for employees, as well as improve on the quality of the rosters, by optimizing manpower employment and deployment costs. Through our contacts, we are aware of some organizations in which a pool of manpower planners spend several days producing each new set of rosters manually, and yet dissatisfactions with the quality of work schedules are routinely expressed by the management and employees. Manpower rostering has been actively researched in the operations research and management science communities for a long time. Popular domains studied in the literature include hospital nurses (Kostreva and Jennings, 1991 ; Miller et al., 1976; Smith and Wiggins, 1977; Warner, 1976), bus drivers (Martello and Toth, 1986; Rousseau, 1984), telephone operators (Henderson and Berry, 1976; Keith, 1979), and airport ground crew (Chew, 1991 ; Holloran and Bryn, 1986). The broad spectrum of domains has led to several distinct solution approaches in the literature, each tailored to a particular class of problems. However, comparatively little work has been done on general models. In this paper, we present R O M A N , a generic toolkit for manpower rostering, developed at the Information Technology Institute in Singapore. The toolkit is a culmination of a year-long applied research effort. Unlike extant systems, R O M A N is designed to address a very wide spectrum of work policies encountered in service organizations. Techniques originating from artificial intelligence, operations research and software engineering arenas are combined into a toolkit that is highly flexible, robust, efficient, customizable and extensible. To date, the toolkit has been licensed to major service organizations in the local health care and transportation sectors. This paper discusses R O M A N in terms of its architectural structure and issues related to its industrial implementation. Analytical modelling and algorithmic aspects in R O M A N will be reviewed only briefly here, since these aspects have been explained in depth in previous publications of the authors (Chew et al., 1991 ; Khoong, 1991 ; Khoong, 1992; Khoong and Lau, 1992). The rest of this paper is organized as follows. In the next section, a generalized view of manpower rostering models is developed. This provides the basis for the discussion of the R O M A N framework in the