Improvements within online decision support in a retail context - Estimation of customer's preferences over freshness of a product in an inventory model with complete upwards and downwards substitution
MetadataShow full item record
This master thesis aims at the estimation of customers' preferences regarding the remaining shelf-life of products available on shelves of a store in a retail context. The research questions focused on how such preferences can be modeled within a stochastic inventory system, how parameters of the demand of such model can be estimated and what is the impact on profit, fill-rate and waste of this model in relation to other similar models. Therefore, an inventory model considering stochastic demand both on quantity and whether the demand is satisfied by the newest items and the oldest items was developed. Perfect downwards and upwards substitution was considered in the inventory model. In this master thesis, upward substitution means that the excess demand for new items can be satisfied by old items and downward substitution means that the excess demand for old items can be satisfied by new items. In addition, the demand was considered to be censored by stock-level. The inventory model developed with a basic base stock replenishment policy generated the data that were used for the estimations of referred preferences which were performed by the adoption of the expectation-maximization algorithm with the support of the maximum likelihood estimation method. To calculate the impact of the estimations on profits, fill-rates and waste; the inventory model with a stock-age dependent replenishment policy was adopted in a design of experiment to compare the use of the estimations acquired from different scenarios. The different scenarios corresponded to variation on mean total demand and how the demand was distributed over the remaining shelf-life of products available including the considered most relevant distribution found in the literature. Among other results, the outcomes showed mainly that in a scenario with both delivery time and ordering frequency equals to $1$ period, the estimations resulted in accurate results. The use of estimation did not show any outstanding general results in relation to other depletion policies. However, the application of the estimations with a stock-age dependent replenishment policy indicated that it has potential to achieve higher improvements on the replenishment of inventory upon the correct calibration and most probably considering higher delivery time and ordering frequency.