Assortment planning framework with substitution and complexity cost
Item Usage Stats
While increasing product variety may have a positive eﬀect on market share, it may lead to diﬃculties in product management and forecasting, leading to increase in inventory holding costs. In addition, as a result of increased setup times, eﬃciency decreases. In this study, we propose two methods to help a multi-national tire manufacturer manage their assortment to ﬁnd a balance be-tween variety and sales. The ﬁrst method evaluates the marginal complexity cost of a set of new products so that a data-driven decision can be made to introduce or not to introduce new products. The second method determines the set of tires to be included in the assortment that increases the total proﬁt. To account correctly for the partial sales revenue losses due to discontinued products, we estimate the fractions of demands substituted by the products left in the assort-ment. Customer no-purchase information were unobserved, as well as the exact timings of stock-outs and sales up to those times. Also, market share information for a particular group of tires for the company were unavailable. Based on these incomplete data, the substitution probabilities were estimated using an iterative method. The solutions were iterated on to ﬁnd the set of substitution probabilities that best ﬁt the data. Discontinued products are expected to save from complex-ity costs due to production capacity losses because of frequent break times. The complexity cost is a set function that accounts for the interactions between the products during manufacturing processes, as well as the variety in the product portfolio. In order to estimate the break time savings, a machine learning model was used, and an algorithm was designed to measure the eﬀect of a set of products being discontinued. For the periods in which the machines work in full capacity, the additional proﬁt due to potential new sales was also considered. This value and the conversion cost saved due to discontinued products were added to the proﬁt function. The predictions from the machine learning model and the other costs are used to formulate a large-scale assortment optimization problem with a complex objective function. The assortment problem is solved using genetic algo-rithm. The results show that the new assortment obtained through our analysis has between 3.8% and 15.2% less products than the initial assortment. The new assortment leads to additional proﬁt between 0.6% and 4.6% of the company’s annual income. The results also show that considering complexity costs in assort-ment decisions leads to substantially diﬀerent assortments and additional savings in comparison to those obtained without their considerations.
Stock-out based substitution
Data driven decision making