Interactive computer aided shift scheduling - XIMES

,QWHUDFWLYHFRPSXWHUDLGHGVKLIWVFKHGXOLQJ -RKDQQHV*$(571(5 ;,0(6*PE+$9LHQQD$XVWULD 9LHQQD8QLYHUVLW\RI7HFKQRORJ\ 6FKZHGHQSODW]±$9LHQQD $UJHQWLQLHUVWU±$9LHQQD This paper starts with a discussion of computer aided shift scheduling. After a brief review of earlier approaches, two conceptualizations of this field are introduced: First, shift scheduling as a field that ranges from extremely stable rosters at one pole to rather market like approaches on the other pole. Unfortunately, already small alterations of a scheduling problem (e.g., the number of groups, the number of shifts) may call for rather different approaches and tools. Second, their environment shapes scheduling problems and scheduling has to be done within idiosyncratic organizational settings. This calls for the amalgamation of scheduling with other tasks (e.g., accounting) and for reflections whether better solutions might become possible by changes in the problem definition (e.g., other service levels, organizational changes). Therefore shift scheduling should be understood as a highly connected problem. Building upon these two conceptualizations, a few examples of software that ease scheduling in some areas of this field are given and future research questions are outlined. ,QWURGXFWLRQ Shift scheduling is important for employees (health, well being, income) as well as for employers (coverage of staffing needs, operating hours, productivity, and costs). Unfortunately scheduling is very complex. The time for solving such a problem soars with exponential speed when the problem size increases (Musliu 2001). Furthermore, many of the requirements are ill defined (Gärtner 1992). An impressive number of researchers worked on these issues. To name a few: Basic mathematical calculi for schedules were discussed in e.g., (Knauth et al. 1982). Deterministic and heuristic strategies for the search process for schedules are described in (Schwarzenau et al. 1984). A further development towards more complex shift-systems is outlined in (Knauth 1987). In (Nachreiner et al. 1993) a system that supports not 'only' the design of rosters but also the design of shifts. Software and algorithms aiming at the evaluation of rosters were developed by e.g., (Jansen 1990; Schönfelder 1992). In addition to highly automated algorithms that still flourish (Musliu 2001) a second approach become at least as important: participation and various aspects of user involvement. This moves towards participation and social issues can be found in scheduling in general (Mc Kay et al. 1989; Hsu et al. 1993; Egger and Wagner 1992), and also in the field of computer support for shift scheduling (Gärtner 1992; Gärtner and Wahl 1997; Gissel and Knauth 1998; Bauer 1999). Not only the research community of the symposia on Night- and Shiftwork addressed issues of computer aided shift scheduling. Further contributions came from Management Science (e.g., Heller et al. 1973; Glover and McMillian 1986; Koop 1988), Operations Research (e.g., Kostreva and Jennings 1991; Ikegami 1996) and Artificial Intelligence (Smith and Bennett 1992) and often aimed for optimal resource utilization. Most of the actual techniques used in these algorithms were developed in the fields of production scheduling, military applications, space, and semi-conductor production (see the good overviews in Tien and Kamiyama 1982; Zweben and Fox 1994). Outside the research community commercial packages aim for non-cyclic scheduling of irregular hours (Blue-pumpkin, ATOSS, Astrum, TempoSoft to name a few out of more than 100). The support for the actual scheduling process stretches from accounting to highly sophisticated algorithms like e.g.

tabu-search and genetic algorithms. In spite of these substantial efforts, the most important tools for shift scheduling probably are self-developed spreadsheets and (small) databases still. Given the long tradition of research and SW-development one wonders why there are so many packages on the market, and why in spite of all this efforts still spreadsheets and databases play such an important role. This paper tries to answer this puzzling fact by two conceptualizations of the field of computer supported shift-scheduling. The first conceptualization discusses features of various shift-scheduling problems as they arise in various industries and companies. The line of argument is that these features of rosters call for different approaches and thereby different tools. The second conceptualization discusses shift scheduling as problem that is highly shaped by its environment (far beyond features of the roster). In this perspective, issues of connecting and integrating come to the fore. This conceptualization suggests that the specifics of the particular situation be of crucial importance to the actual scheduling process and that the corresponding idiosyncrasies cause the difficulties in developing proper tools. These idiosyncrasies increase the complexity enormously. How to manage such complexity within the changing field of shift-work (towards more irregular hours) is in this perspective the central question to be addressed in the next future. 7KHILHOGRIVFKHGXOLQJ This section deals with four issues. First, I conceptualize a diverse field of shift systems by describing two poles. Second, I argue why these differences are important to the scheduling process and thereby to the design of corresponding tools. Third, I give examples of software that address specific areas of this field. Fourth, I discuss first consequences of this diversity. )HDWXUH 3ROH3DSHUPLOOV 3ROH0DUNHWV Shift groups A few rather stable groups No stable groups Shift patterns Regular Regular & irregular Number of shifts Small No stable shifts Number of employees Fix Changing Staffing levels at diff. times Constant Varying Role of different qualifications Considered when building groups To be scheduled Staffing needs Precisely defined A probability distribution Predictability of needs High Low Planning horizon Weeks & months Starting with hours Individual preferences Hardly considered Considered strongly Spatial constraints Not considered Considered strongly Table 1: Two poles illustrating the diversity of the field of shift scheduling It would be wrong to think of this field described in Table 1 in terms of two approaches to shift scheduling. There are these two approaches and many, many in between. The importance and appearance of each single feature of the above table may make a substantial if not decisive difference to the scheduling approach (exploiting different simplification strategies to minimize computational complexity) and to the tools to be used (e.g. calling for very different user interfaces). Scheduling problems vary in the features listed above and in others. Even paper mills are not always as stable as the use in the table above might suggest. The notion of ’paper mills’ was only used here to illustrate the pole. Variety exists in other industries too, often in contrast to presumptions (e.g., stable teams in call-centers, extremely flexible work for maintenance in steel mills, stable time patterns in spite of flexi-time). To add to the complexity, the selection of planning strategies goes beyond technical decisions. Different scheduling techniques may reflect different optimization approaches. One approach may focus on minimizing the load caused by work (often to be found in 'male-industries' – e.g., heavy industry). Another approach may focus on minimizing all loads (work, travel, family,… ) of-

ten to be found in ’female industries’ (e.g., health). Furthermore issues of industrial relations (e.g., self determination versus collective bargaining), the role of laws, political fairness (strong versus weak groups on the labor market), importance of informed decision making (e.g., can workers assess the risks they take, are there effects on third parties) influence the selection of planning strategies. To illustrate the level of development achieved so far and the diversity of tools needed to cover this inhomogeneous field I discuss two systems shortly. The first system (ShiftPlanAssistant – SPA, developed by our group) facilitates the design of somehow regular rosters. This ranges from 'paper mill' problems to rosters for a small number of individuals with not too many shifts, for cyclical and - up to some degree - non-cyclical rosters without too much care on qualifications or individual preferences, spatial requirements, etc. In Figure 1, the top area shows the roster for several individuals. The middle area shows whether staffing requirements are met (they are met = there are no differences). In the lower part information on time-accounts is given. Various additional features allow for generation of rosters, legal checking, ergonomic analysis, cost assessment, printouts, etc.

Figure 1:

ShiftPlanAssistant (SPA) used for a somehow irregular problem

A rather different system is shown in Figure 2 (TimeCare, Sweden). One core idea is to use no shifts anymore. Individuals may select time frames of work freely. In regular intervals (e.g., several weeks) remaining under- and overstaffing outside the limits is straightened out. This is done by a prioritization strategy. Qualifications are not considered too strongly in the version shown below. However, the group declared qualifications to be a focus of further development. This system is much closer but still different to the market pole (e.g., a well-defined clearing process is necessary every few weeks; a well-defined set of employees is involved). Another interesting features (also closer to the market pole) is that it exploits differences between minimum and maximum staffing levels (theses levels are illustrated with the blue and red lines). If such differences are attainable, they allow for more freedom to select duties. However, they can not be found in every occupation.

Figure 2:

TimeCare for scheduling without shifts

Above, I tried to exemplify the diversity of shift scheduling. The consequences of this diversity are substantial. If software tries to cover large areas of this field then the software may become too complex to be used by many users (especially if the software is used seldom what can be expected in many cases). However, if several small packages are developed that aim for small areas, it may, again, become very complex to find the appropriate package. Meta-strategies are needed. &RQQHFWLQJ LQWHJUDWLQJ In the above chapter the focus was on the design of tools for designing rosters. This chapter conceptualizes a broader field: How can VKLIWVFKHGXOLQJ SUREOHPVEHFKDQJHG to allow for better rosters and what other issues are relevant in this process. Consider the following roster:

Figure 3:

An example for a tight scheduling problem (M12...Morning Shift 12h; D…Day; Red …vacation; Sick …coverage of sick leave; Amb …being on call).

At least four factors make it difficult to develop good (or at least reasonable rosters) for this problem: average working hours are high; absence causes many additional duties; there is less work

on Saturday; the time critical character of work makes it necessary to have stand-by duties. In order to improve the roster one has to go beyond rearranging duties. A transformation of ’the problem’ has to be considered (e.g., changes in the organization of work, reduction of working hours). A further expansion of the issue of problem setting (i.e. how problems are bordered in the sense of what to consider as changeable and what has to be considered in the course of optimization, compare Gärtner 2001) is crucial. Optimization of rosters may be only a part of several relevant scheduling and none scheduling issues. Such additional issues may shape the software as well as the actual rostering process. Table 2 lists examples for such additional issues. ,QWHJUDWLRQRIVKLIWVFKHGXOLQJ ,QWHJUDWLRQRIVKLIWVFKHGXOLQJZLWK ZLWKRWKHUVFKHGXOLQJWDVNV QRQVFKHGXOLQJWDVNV ♦ forecasting ♦ the organization of work (who does what when) ♦ route planning & vehicle planning ♦ managing customer data ♦ team structures for better customer service (e.g. health care) ♦ accounting ♦ considering special qualifications ♦ www access to calendar data ♦ considering limits caused by resources (machines) ♦ ... ♦ … Table 2: Other issues that may shape shift scheduling processes and tools These additional issues may influence the ergonomic quality of rosters strongly but may, again, be very complex. An example for a software developed by our group supporting one such task, the design of shifts and the computation of staff needed, is shown in the next figure.

Figure 4:

A screen shot of the system Operating Hours Assistant (OPA) that facilitates the analysis of operating hours, staffing needs and the design of shifts.

The screenshot of the system shown in Figure 4 displays shifts generated to meet the staffing needs visualized in the diagram. In general, the system helps to fit staffing and staffing needs, to design good shifts (good start/end times; not too long/short duties), to keep the number of duties per

week below critical thresholds (e.g. approx.