| الوصف: |
As a consequence of technological advances and product customization, manufacturing companies have considerably increased their product portfolios over the last decades, resulting in high complexity. This has led to low product availability, low accuracy in demand forecast, and high levels of obsolescence at Hilti. Therefore, the objective of this project is to fight against this complexity by developing an algorithm that classifies products into "low performance" or "good performance", in order to subsequently phase-out, replace the product with a new one in the market, or simply change the sourcing strategy. The focus of the work will be only the development and creation of the algorithm. For its construction, first of all an intuitive algorithm is elaborated. Then, to test its validity, a total of 685 products are evaluated thanks to interviews with Product Managers, where a classification of "low performance" or "good performance" is obtained for each of the assessed products. Once the labels have been obtained for each product, two other traditional algorithms are developed and their results are examined. Then, two Artificial Intelligence algorithms in the form of Decision Trees are constructed to improve the results of the first two. The main focus is to evaluate how these models perform when tested with products from the same family, and how accurate they are when scaled to other product families. In addition, it is intended to use variables that take into account the relationship between products, and to answer the following question: Do Artificial Intelligence algorithms improve the results of traditional algorithms when it comes to performance evaluation in product portfolio optimization? The results of the work confirm that indeed Artificial Intelligence algorithms improve the results of traditional algorithms, not only when they are tested for the same family of products, but especially when they are escalated to other families of items. In fact, the improvement from traditional algorithms to the final Artificial Intelligence model is 181% in precision. Moreover, the importance of linked revenue variables for the algorithms development is confirmed. Finally, the impact that the implementation of such algorithms could have on Hilti AG is studied. In this case, the implementation would cause a reduction of 62% in obsolescence, in addition to an 8.4% reduction in storage volume, having a maximum risk of a negative decrease of only 0.7% of the sales volume |