Forecasting Incoming Calls to Telemarketing Centers

Many studies in the past have demonstrated the usefulness of Holt-Winters' multiplicative forecasting model to be the best for forecasting items such as revenues, demand, earnings per share and inventory and stock levels whenever the time series data contained elements of seasonality. Holt-Winters' model is an extrapolative technique which isolates the level, trend, and seasonal components before generating forecasts regardless of the nature of the time series. It provides a cost effective alternative to more sophisticated methods such as ARIMA modeling of the Box-Jenkins variety. Furthermore, it is more effective and cost efficient than using so-called "expert opinion" in most cases. Telemarketing centers are responsible for answering incoming requests from customers who desire immediate attention. Customers who are placed on hold are least satisfied. The economic losses to firms, whose telemarketing systems place large numbers of incoming calls on hold, can be great in terms of the loss of incoming revenue and future economic opportunities. Employees of the center feel most satisfied when the number of calls on hold are reduced to a bare minimum. Less customers on hold means fewer unpleasant customers. Managements satisfaction is also the greatest when revenues are maximized and costs are minimized for servicing the incoming calls. Through forecasting and planning, management can reduce the number of customers on hold. Currently, ATT 314 (Missouri) and 504 (Louisiana) from the Central Time Zone; 303 (Colorado) from the Mountain Time Zone; and 808 (Hawaii) from the Pacific Time Zone. The data used were of actual daily call volumes from March 1, 1991 to June 26, 1991. Initial plots of the data revealed weekly seasonality, the presence of outliers, and no apparent trend over time. As an example, the day to day behavior of call volumes within area code 201 (Figure 1)reveals two outliers; one on March 14, 1991 and the other on May 6, 1991. (Figure 1 omitted) Also, there is a pattern during the week, with Monday having the largest number of calls and Sundays the least number of calls. (See Figures 2 and 8) Close inspection of the data reveals other days which may be considered outliers as well. ( Figures 2 and 8 omitted) In the same figure additional outliers can be found during May, 1991. Figure 3, a similar time plot for Tuesday calls does not contain any outliers nor does Figure 4 (of Wednesday), Figure 7 (of Saturday) and Figure 8 (of Sunday). (Figures 3, 4 and 7 omitted) There is one outlier in Figure 5 (of Thursday) and possibly two (or none) in Figure 6 (of Friday). (Figures 5 and 6 omitted) Thus, the total number of possible outliers during the period studied is five and could be reduced easily to three since two of the outliers are not very different from the patterns observed in the time series of daily incoming calls of the area code studied. In examining the time plots, we find that the majority of the outliers appear close to holidays; for example, the outlier of March 14, 1991 is close to St. …