How Does L.L. Bean Use Past Demand Data and a Specific Item Forecast to Decide How Many Units of That Item to Stock?

...

the book forecast and the item forecast, the item forecasts need to be adjusted to stay in line with the overall book forecast.

The information Scott needs includes the market and fashion trends that affect shirt demands for the new shirt and how that affects other shirts in the catalog, the expected dollar values and the book forecast in order to arrive at a demand forecast for the new item.

4. How would you address Mark Fasold’s concern that the number of items purchased usually exceeds the number forecast?

Mark Fasold is concerned about the dispersion in the forecasted errors for both “new” and “never out” items. He was also concerned that the methodology gave way to excess inventory orders because the underage cost was greater than the overage cost. One effective way to deal with this is to first identify the costs involved in the optimal probability fractile correctly. In order to correctly identify the overage and underage, the per-item cost to the retailer has to be correctly calculated to include inventory handling and shipping costs, and even the retail store cost by store occupancy and should be included in the cost-benefit analysis before deciding to order greater than the frozen demand forecast.

The errors are widely dispersed, and one reason could be that all forecast errors are compiled according to new and never out, although these categories vary largely across L.L. Bean’s product offerings. For example, a new men’s shoe from the past year, will affect the forecast for a new women’s earring, creating large discrepancies. This has to be better structured. One way of doing this is by identifying smaller clusters of items to compile and calculate forecast error frequency distributions. This can be done by first clubbing similar items and then clubbing items that usually affect the sales of each other – for example women’s athletic shoes and women’s socks can be under the same cluster, whereas including men’s ties will not help. The dispersion will likely reduce, therby reducing the the demand distribution optimal fractile value, or more accurately predicts the overstocking need.

Third, preferring vendors with a shorter lead time or incentivizing vendors with longer lead times to deliver faster will create a faster turnaround period. Giving L.L. Bean the opportunity to analyze actual demand for the item post introduction, and to accordingly restock their inventory instead of overstocking right from the beginning. Also since new items are overstocked more, adding them to more relevant catalogues will give L.L. Bean the opportunity to get more exposure to the product and might push up demand for the item.

5. What should L.L. Bean do to improve its forecasting process?

Apart from the above-mentioned methods to improve forecasting, L.L. Bean can incorporate several other changes. L.L. Bean should have a more iterative process of forecasting, and should continually update the forecast errors information and the frequency distribution to catch trend spikes and changes, this can be done, for example, weekly instead of quarterly. Since external factors such as changing fashion trends have a huge impact on demand and items change more often and faster in the fashion industry, L.L Bean can employ some fashion experts to continually monitor market trends and to report to the respective teams. To pre-empt wrong forecasts, catalogs can go out a little earlier than product launch and customers can be incentivized for placing pre-orders on items. New items can be compared with never outs and the cost of overstocking never outs that are no longer in demand can be reduced.

...