As we discussed at the beginning of the chapter, Starbucks has a large global supply chain that must efficiently supply over 17,000 stores. Although the stores might appear to be very similar, they are actually very different. Depending on the location of the store, its size, and the profiles of the customers served, Starbucks management configures the store offerings to take maximum advantage of the space available and customer preferences. Starbucks actual distribution system is much more complex, but for the purpose of our exercise lets focus on a single item that is currently distributed through five distribution centers in united states. Our items is a logo- branded coffeemaker that is sold at some of the larger retail stores. The coffeemaker has been a steady seller over the years due to its reliability and rugged construction. Starbucks does not consider this a seasonal product, but there is some variability in demand. Demand for the product over the past 13 weeks is shown in the following table.
The demand at the distribution centers (DCs) varies between about 40 units on average per week in Atlanta and 48 units in Dallas. The current quarter's data are pretty close to the demand shown in the table. Management would like you to experiment with some forecasting models to determine what should be used in a new system to be implemented. The new system is programmed to use one of the two forecasting models: Simple moving average or exponential smoothing.
Week |
1 |
2 |
3 |
4 |
5 |
6 |
7 |
8 |
9 |
10 |
11 |
12 |
13 |
average |
Atlanta |
33 |
45 |
37 |
38 |
55 |
30 |
18 |
58 |
47 |
37 |
23 |
55 |
40 |
40 |
Boston |
26 |
35 |
41 |
40 |
46 |
48 |
55 |
18 |
62 |
44 |
30 |
45 |
50 |
42 |
Chicago |
44 |
34 |
22 |
55 |
48 |
72 |
62 |
28 |
27 |
95 |
35 |
45 |
47 |
47 |
Dallas |
27 |
42 |
35 |
40 |
51 |
64 |
70 |
65 |
55 |
43 |
38 |
47 |
42 |
48 |
LA |
32 |
43 |
54 |
40 |
46 |
74 |
40 |
35 |
45 |
38 |
48 |
56 |
50 |
46 |
Total |
162 |
199 |
189 |
213 |
246 |
288 |
245 |
204 |
236 |
257 |
174 |
248 |
229 |
222 |
1) Consider using a simple moving average model. Experiment with models using five weeks abd three weeks past data. The past data in each region are given below( week -1 is the week before week 1 in the table, -2 is two weeks before week 1, etc.) . Evaluate the forecasts that would have been made over the 13 weeks) mean absolute deviation, mean absolute percent error, and tracking signal as criteria.
Week |
-5 |
-4 |
-3 |
-2 |
-1 |
Atlanta |
45 |
38 |
30 |
58 |
37 |
Boston |
62 |
18 |
48 |
40 |
35 |
Chicago |
62 |
22 |
72 |
44 |
48 |
Dallas |
42 |
35 |
40 |
64 |
43 |
LA |
43 |
40 |
54 |
46 |
35 |
Total |
254 |
153 |
244 |
252 |
198 |
2) Next, consider using a simple exponential smoothing model. In your analysis, test two alpha values, .2 and .4. Use the same criteria for evaluating the modle as in part 1. When using an alpha value of .2, asume that the forecast for week 1 is the past three week average( the average demand for period -3, -2, and -1). For the model using an alpha of .4, assume that for week 1 is the past five weeks average.
3) Starbucks is considering simplifying the supply chain for its coffee makers. Instead of stocking the coffeemakers in all five distribution centers, it is considering only supplying it from a single location. Evaluate this option by analyzing how accurate the forecast would be based on the demand aggregated across all regions. Use the model that you think is best for your analysis of parts 1 and 2. Evaluate your new forecast using mean absolute deviation, mean absolute percent error, and tracking signals.
4) What are the advantages and disadvantages of aggregating demand from a forecasting view? Are there other things that should be considered when going from multiple DCs to one DC7.