LL Bean Essay

1. How significant (quantitatively) of a problem is the mismatch between supply and demand for LL Bean? As per the historical series and its associated statistical description (see graph below), we can observe that there is a significant spread between the A/F ratios sine the standard deviation equals 1/3 of the mean. Besides in cases, there is mismatch beyond 50% between the forecast and the actual demand. Besides the mean value shows that there is a 9% bias meaning that on average the actual is always 9% above the forecast. It should be noticed as well that there distribution is skewed to the left with higher values meaning that there is a 100% underestimation for certain items. 2. Use the provided Excel file that contains demand and forecast data for a collection of items. Suppose those are the data LL Bean will use to plan their next season. Consider an item that retails for $45 dollars and costs LL Bean $25. The liquidation price for this item will be $15. The sales forecast for this item is 12,000. What order quantity would LL Bean choose for this item? Based on the Cu/(Co+Cu) ratio that equals 20/(10+20) =0,667 and the A/F distribution, we end-up with a probability of 0,676 given the round up rule. Hence LL Bean should order 12 000 * 1,179975 = 14160 items to maximize its profit. (We used the distribution derived from the data rather than the normal distribution with the same mean and standard deviation. Indeed despite the important gaps between the different percentiles of the real distribution, we reject the hypothesis that the distribution is normal at a 5% level as per the Anderson Darling test result with p-value= 3%). 3. Assuming LL Bean manages to derive the correct forecast, what do you think about their ordering process? (You may wish to begin with Mark Fasold’s concerns at the end of the case. Also, think about Rol Fessenden’s concern about estimating contribution margin and liquidation costs). †¢ If the contribution margin and liquidations costs are wrongly assessed this has a direct consequence on the commitment order size as per the newsvendor model methods (cf. the Cu/(Co+Cu) ratio). †¢ There is a grey area in the case to know how LL Bean really assesses the number of actual for products generating a demand higher than the forecast. An overestimation of lost sales can create a bias loop since it will impact the next year order commitment by generating mechanically higher commitment orders. As per the mean (8% above 1) and the distribution that is skewed to the left, it could be inferred that there is a systematic overestimation of lost sales which may explain that there are not different common pattern across items and buyers. †¢ We can’t suggest any bias due to outlier since they mention that there have not found any specific pattern. †¢ The split between â€Å"new† and â€Å"never out† for the historical errors makes sense since both nature of articles share a common property. †¢ We recommend making use of the phone calls and orders through all selling channel to build more robust analytical data and reduce the potential bias of data used to build the A/F distribution. 4. What do you think about LL Bean’s forecasting process? Is that the best that they can do? Problems †¢ It seems unreliable and not data driven as per the use of rules of thumb and use of consensus that may reduce the weight of the expert. †¢ Forecast reconciliation issue with the bottom up (items by items) and the top down (catalog) approach forecast approach. †¢ A lot of the forecast relies on the inaccurate slash at the end of the process. †¢ Aggregation of demand for item common to different catalogs seems unclear and prone to error, there may be an overestimation of the demand forecast by double counting the expected sales (cf. catalog arriving to same customer that are considered the best i.e. buying the most). †¢ Issues with the impact of new products and cannibalization †¢ Differences observed between the aggregation Suggestions †¢ More frequent interactions between bottom up and top down approach to avoid or at least reduce the slash of the end. Such interactions could be achieved through the so-called â€Å"W† approach that implies meeting points at different levels over the process. †¢ For items common to several catalogs, consider a customer approach instead of a catalog approach to avoid counting several times the expected purchase of one customer receiving several catalogs. †¢ We recommend making use of the phone calls and orders through all selling channel to build more robust analytical data in order to improve the forecasting process. †¢ Try to find alternate sources of supply to reduce the current lead time of 9 months and allow finalizing the forecasting process closer to the sales time.