13.1 13.2 13.3 13.4 13.5 13.6 13.7 Introduction Quantitative Forecasting Causal Forecasting Models Time-Series Forecasting Models The Role of Historical Data: Divide and Conquer Qualitative Forecasting Notes on Implementation
KEY TERMS SELF-REVIEW EXERCISES PROBLEMS CASE 1: BANK OF LARAMIE CASE 2: SHUMWAY, HORCH, AND SAGER (B) CASE 3: MARRIOTT ROOM FORECASTING REFERENCES
CD13-2 C D
C H A P T E R S
Forecasting Improvement at L. L. Bean
increase dramatically. On the other hand, overstaffing of either group of agents incurs the obvious penalty of excessive direct labor costs for the underutilized pool of agents on duty. The staff-scheduling decisions would be quite routine if it were not for the erratic nature and extreme seasonality of L. L. Bean’s business. For example, the three-week period before Christmas can make or break the year, as nearly 20% of the annual calls come during this short period. L. L. Bean will typically double the number of agents and quadruple the number of phone lines during this period. After this period, there is, of course, the exact opposite problem, the build-down process. In addition, there is a strong day-of-week pattern throughout the year in both types of calls, with the volume in the week the highest on Monday and monotonically decreasing down to the low on Sunday. Other factors that must be considered by the forecasting model is the effect of catalog mailings or “drops.” These are generally done so that the bulk of the catalogs arrive around Tuesday, which disrupts the normal pattern of calls tremendously. Many eager customers order immediately, which creates a surge of new calls around the time of the “drop.” The new forecasting model that was developed had much greater forecast accuracy than L. L. Bean’s previous approach and was able to produce a mean absolute percentage error of 7.4% for the TM group and 11.4% for the TI group on five years of...