Content area
E-commerce is a rising trading manner that is driving customers towards an increasingly demanding behaviour. Since e-customers still perceive the shopping activity as uncertain, information sharing is a vehicle to customer trust and hence customer retention. Sharing delivery status and estimated delivery dates with customers enhances their satisfaction and their repurchase intention.
The current project seeks to estimate delivery dates in a multi-brand luxury fashion e-seller. This delivery process spans several entities and stages, starting in worldwide scattered boutiques until final customers. As such, this project comprises six independent sub-problems that represent the order processing phases. For each one of these, a predicting model was created to determine the corresponding timespans, based on a range of independent factors that ought to characterize each specific order. This was supported by several data mining techniques such as data cleaning, classification and regression. Data cleaning and classification were performed in order to reshape data so that modelling results could be obtained or improved. Data cleaning mainly consisted in outlier removal and variable standardization, while classification’s purpose was to decrease the number of levels of certain categorical factors. Modelling was an iterative process in which different techniques were explored, according to the previous modelling results. The main tool used in this stage were decision trees that created distinct order groups based on the combination of the chosen independent factors.
Due to the fact that each sub-problem comprises different processes and data, models’ results diverged. With the exception of step 6 (in transit), results were satisfactory, which was evaluated based on several mean error measures. Concerning this step in particular, due to its importance in the scope of the delivery, factor route should be treated with more detail in order to improve model performance. Overall, expected timespan measurements were successfully obtained for each combination of order factors, which also constitutes a meaningful insight on what impacts on delivery performance. As this is an iterative process, once data is updated, other teams can be involved in the project in order to implement this tool. This is expected to positively impact on both business control and customer satisfaction.