Sales Forecasting: A "Job Shop" Case Study Revisited

Maxwell K. Hsu, Tennessee State University P. Byron Pennington, Tennessee State University Festus Olorunniwo, Tennessee State University

Abstract Using a twenty-four (24) month period of single-dimensional sales data reported in a recent published article, our analysis shows that 'Marketers should not solely rely on a commercial forecasting package to make their forecasting estimation. Indeed, small and mid-size business managers can achieve significant savings by wisely utilizing their educated business experience along with a spreadsheet software (e.g., EXCEL) or a forecasting software (e.g., Focus Forecasting Package). Introduction Many medium and small businesses have traditionally shied away form using forecasting techniques due to the perception that these techniques are expensive and too difficult to use. in their recent article, Muscatello and Coccari (hereafter referred to as "M&C") (2000) presented a simple case study of a job shop manufacturer competing in the replacement parts market. In that example, the job shop used a "naive" forecasting method in which the forecast for next period (F,,,) is simply the sales for the current period (S,). Such naive forecasting (F, = S,, t =1; F,,, = S, t-- 2, 23) takes no account of seasonal, industry, or economic cycles typically characterizing market sales/production data. Simply, the "naive" forecasting was based solely on last month's sales. Using real demand data from the replacement parts manufacturer, M&C argued that a formal forecasting is much better than the abovementioned "naive" forecasting approach and they concluded that "enormous (cost) savings ... can (be) achieve(d) by using a simple (formal) forecasting technique" (2000, p. 21). In order to reduce forecasting errors, M&C relied upon the Focus Forecasting Package to select the best fitting forecasting technique that produces the optimal outcome, using the historical sales data from the job shop. The Focus forecasting package suggested the best model with seasonality and an alpha of 0.5 to match the data. M&C (2000) were able to show that results produced by the optimal Exponential Smoothing Approach ("ESA") is superior to the "naYve" forecasting approach on the basis of mean absolute percentage error (MAPE) with considerable cost savings.

In their example. 1996). From the modem marketing standpoint. Accordingly. Armed with this knowledge and their capability to use spreadsheets. they assumed that all excess inventory was sold off at the end of each month with a holding cost of $20 for each overproduced item. p. On the other hand. the conventional wisdom suggests that it is far more expensive to lure a new customer than to keep a satisfied current customer (Reichheld. Obviously. One is referred to as a low season that covers month twelve (12) as well as months one (1) through seven (7) with relatively lower sales (average is 431 units per month).We become interested in M&C's (2000) study for the following reasons: 1) The so-called "superior" ESA under-forecasted the sales by 236 units which amounts to an equivalent loss of $42. Indeed. The same pattern was observed for the second year.480 ($180/unit x 236 units = $42.480) over the subject 24 month time period. 3) Most managers have deep knowledge of their businesses. 2) Most off-the-shelf software products may not have the capability to handle simple scenarios specific to each individual business. M&C's (2000) analysis implicitly reflects such a notion. each under-produced item represents not only a lost sale but also a potential lost of one loyal customer. the job shop is in a business with two unequal seasons and its sales forecast could be improved if we use data available in the first twelve months to project sales in the following year. In fact. Taking the studied job shop for example there are only two seasons in a year (see Table 1). each underproduced item was assumed to represent a loss of $180. The other is referred to as a high season represented by months eight (8) through eleven (11) with relatively higher sales (average is 498 units per month). 38). it is likely that they may be able to produce better forecasts by using simple forecasting methods (instead of off-the-shelf software products) that are geared toward the specific nuances of their businesses. though over-production leads to excess inventory and increased inventory cost. 1999. . a case with an acceptable over-production is better than stocking-out with insufficient production. It is well established in the concept of marketing orientation as well as the definition of "business" that the firm and its customers "make a comnitment to each other to do business over a long period of time rather than view each sale as a discrete transaction" (Stanton and Spiro.

small and mid-size businesses could utilize educated common sense to further improve M&C forecasting results. we demonstrate how managers could take advantage of their business experience and any widely used spreadsheet software (e.The present paper attempts to show that. In addition. using data from the same case study reported by M&C (2000). We build the MESA forecast on the basis of M&C's ESA results.g. "ESA".e. represented by negative errors) and under. etc.) to produce satisfactory forecast results. which reflects business managers' willingness to tackle the more costly under-production problem. In addition. which will be discussed shortly) in producing the sales forecast for the second year (i. Works. Lotus. months thirteen (13) through twenty-four (24)). . we also examine the economic costs of over. Excel. MESA and SWA.. and the "SWA"). "MESA". We then compare the forecasting errors between our proposed methods with M&C's (2000) ESA. "Naive". we decide to use two simple methods (i.. Forecast The first method is referred to as a modified ESA ("MESA") approach.estimation/under-production (forecast number was less than the actual.e.It is worthy to note that error is defined as "actual sales minus forecast sales" (M&C...e. Proposed Forecasting Methods Since the job shop manufacturer is involved in producing a seasonal product. represented by positive errors). 2000) and an error can result from overestimation/overproduction (forecast number was greater than the actual.and under-production among the four models (i.

the "naive" forecast produces a Poor MAPE score while the proposed MESA forecast produces the best MAPE score among the four competing models. F. the bigger number between actual sales in month seven [398 units) and the M&C's computer-produced ESA forecast for month nineteen [389 units] (see Table 1). In this 'Job shop" case. The second method is referred to as a Simple Weighted Analysis ("SWA") approach that also incorporates judgmental assistance. Following M&C (2000).13 = (S1112 / St)*S t-1.4%. ESA. MESA. 2. . the educated MESA approach beats the oversimplified "naive" practices as well as M&C's complex computer forecasting outcome (see Table 4) Notably. Then we multiply this calculated weight with the actual sales occurred twelve months ago to produce a first-stage predicted sales (i. The advantage of the SWA approach is that it empowers the manager to (1) adapt the method to any changes in the data trend. our proposed SWA approach also outperforms the M&C claimed "superior" ESA forecast in producing a lower MAPE index as well as a lower total cost. the predictive power of the competing models is assessed by one or more of the performance measures.3%. and 3.. By so doing. the MESA forecast would be 398 units. Taking month nineteen for example.. Next. 3. we select the maximum number of one-year-ago sales and the predicted sales to better handle the costly under Production problem.. t = 1. 2000).9%. we first compute a ratio between sales in last month and the sales volume thirteen months ago.e. For month fourteen (14) and after. Even when forecasting software such as the Focus Forecasting Package is not available.4%. 2. That is. Radobilshy and Eyck.Specifically. the MESA predicted sales value would be the larger number between the actual sales one year ago and the ESA forecast. we direct our attention to the cost analysis. and (2) use his/her knowledge base of the operation to produce the forecast that best suits his/her own business. respectively (see Table 4). we compare the predictive power of the four competing models using the MAPE. Similar to OUT proposed MESA approach. We use actual sales in month one (1) as a proxy to represent forecast sales in month thirteen (13). Typically. MESA and SWA).). and SWA forecasts are 7. Similarly. . The MAPE index for the Naive. for instance. we link the computer-forecasted results with actual sales a year before to further reduce the significance of under-production problem. This procedure could easily be handled by using spreadsheet software such as Excel (cf.. Results Table 2 and 3 report t e recast ng results associated with our proposed methods (i. the final predicted sales value for SWA would be determined by the larger number between the actual sales one Year ago and the SWA first-stage forecast number. 3.e. results show that small and mid-size business managers should take advantage of their educated business experience in producing forecast numbers with the help of a spreadsheet software.

combined with decent knowledge of their business. The over.and underproduction numbers in M&C's tables were the simple totals of each. In other words. Accordingly.Concluding Remarks This investigation shows that. we should note that no lead-time data (daily/weekly) were included in M&C's study as if the production scheduler worked one day each month. small and midsize business managers can achieve significant savings by using a simple spreadsheet application. no absorption period was acknowledged. However. . our analysis is solely based on a twenty-four (24) month period of single-dimensional sales data.

For a longer time-series data. price changes. regression analysis and/or more complicated multi-dimensional VECM (Vector Error Correction Model) approach might prove to be helpful in forecasting. future forecasting study should explore important marketing information such as promotions. .A more complex computer model would definitely be required when two or more variables enter the system. competition. considering industry and business economic cycles. At the firm or industry level. and product innovations.

Profits. Boston. MA: Harvard Business School Press. 1999. MA: Irwin McGraw-Hill. The Journal of Business Forecasting. and Ronald L. William J. 19(1) (2000): 19-2 1.References Muscatello. The Journal of Business Forecasting. Spiro: Management ofA Sales Force. Frederick F. 1996. and Lasting Value. Stanton. 19(3) (2000): 22-27. Joseph R. Boston. Radobilshy. Coccari: Savings with a Forinal Forecasting System: A Case Study. Reichheld.: The Loyalty Ejj-ect: The Hidden Force Behind Growth. and Rosann L. Zinovy and John Ten Eyck: Forecasting with Excel. .