Model Implementation for Aggregate Output Scheduling.
Gene K. Groff · Academy of Management Proceedings · 1966
In summary, the starting point for model implementation should be the problem. In the case of aggregate output scheduling, it is useful to distinguish between planning and control problems, and the associated model use. Planning is concerned with the anticipated workforce and inventory levels for future periods in order that preparatory activities might be accomplished. The control problem is substantially more subtle and complex. This problem involves sequential use of the model and explicit means of dealing with demand uncertainty. It was noted that the proper criterion for developing model parameters, when the models and cost structures are not perfectly matched, should be the expected operating costs and not the goodness-of-fit of the model to the data. Curves that appear to fit the data reasonably well may not result in operating costs as low as those achieved with a more illogical or poorer fit. Substantial cost differences were found between curves established by least-squares techniques and those established by trial and error. It was also noted that the forecast characteristics were important to the model's effectiveness. The forecast bias was found to be especially relevant. Thus, any attempt at model implementation for the aggregate output problem should also be accompanied by a thorough investigation of the forecasting system. The costs associated with holding buffer stock for the peak period should not be assumed to be a single month's holding charge. The cost will obviously differ for each problem; but, in this investigation, it was found to be much greater than a single month's holding cost. Finally, a twelve-month planning horizon and a frequent replanning frequency were found to be desirable. The problem of cost estimation is both important and difficult. If a reasonable basis for investing in a new aggregate output scheduling procedure is to be established, operating costs must be estimated; the operations researcher's credibility, and thus his long-run effectiveness, is also involved. Unfortunately, the prediction of operating costs is not easy. It was found that the control use costs were approximately 20% higher than the planning model costs. The relative standard deviation of the annual costs of the control model was approximately 10%; thus, even though one could predict the expected penalty costs with some degree of accuracy, the estimate for any one year might involve a substantial error. While the planning model was found to be insensitive to errors in cost estimation, the control use of the linear programming model exhibited greater sensitivity. Finally, if the forecast is not reasonably accurate with known characteristics, the expected cost is particularly hard to estimate. While results from an investigation of a single cost structure can be an unreliable basis for generating sweeping conclusions, they can be used to illustrate the need for careful development of strategies for model implementation. The task of utilizing models to assist in making operating decisions and the prediction of operating results can be a difficult and subtle problem. Model applications are not likely to be outstandingly successful if sound strategies for utilization are not developed, and if both operating managers and operations researchers do not have an adequate understanding of the interactions between the model, the decision strategy, and operating results.