Perishable item inventory management has been studied by researchers in the area for over four decades (e.g., Broekmeulen and van Donselaar, 2009, Cohen, 1976, Fries, 1975, Nahmias, 1975, Nahmias and Pierskalla, 1973, Pierskalla and Roach, 1972, Prastacos, 1981 and Van Zyl, 1964). Bakker et al. (2012) provide an excellent overview of research on inventory systems with deterioration. Managing inventory of perishable items is different from those for items that have a relatively longer shelf-life simply because of additional shrinkage that arise due to unsaleables as well as related dynamics. Regardless, the general dynamics in this area is known to be a good degree of accuracy for reasonably accurate planning purposes. However, given the thin margins faced by retailers that are more often a norm than an exception, any improvement in the accuracy of managing inventory would result in significant tangible benefits in terms of less wastage, more profit, increased customer satisfaction, among others.
Given the higher level of granularity afforded by the identification technology of choice – bar codes – a majority of the literature on perishable item inventory management model and perform analysis at the class level where an instance of this class is not differentiated even though the conditions encountered may not necessarily be identical across different instances. While these suffices as a good approximation, it is far from reality since (1) given the short remaining shelf life of perishable items, the consequences of even minor over- or under-estimation could be profound – it is relatively easy to over- or under-shoot in estimates and (2) perishables are notorious for exhibiting different rates of degradation, resulting in different remaining shelf lives even within the same pallet due to their exposure to different environmental conditions in transit and during storage.
Perishable items are generally maintained in a controlled environment (e.g., refrigeration, low humidity) to slow down their quality degradation rate. However, it is extremely difficult to maintain consistency in ambient conditions even within a pallet (e.g., Jedermann et al., 2010 and Praeger et al., 2012). Reasons for this include the density of surrounding material(s), relative distance from cooling units, etc. Therefore, given that different items are exposed to different ambient conditions, it helps to know as accurately as possible the ambient condition profile over time for each individual perishable item as it passes through the supply chain until it reaches the final customer. Recent developments in sensor and auto-identification technologies such as Time–Temperature Indicators (TTI), RFID (Radio-Frequency IDentification) tags (e.g., Becker et al., 2010, De Marco et al., 2012 and Ngai et al., 2008), among others, facilitate the ease of accomplishing the generation of item-level information. TTI stickers are placed on the perishable object and a change in color in these stickers is used to determine the extent of spoilage of the perishable. However, a drawback of TTIs (vs. RFID) is that these are passive and cannot communicate with a reader. RFID tags, on the other hand, can communicate with a reader for continual status updates.
We consider inventory management of perishables where the demand for an item is determined by its allocated shelf-space as well as its instantaneous freshness. As information at a finer level of granularity (e.g., through item-level RFID tags) become available (e.g., Zhou, 2009), there is a concomitant need to be able to utilize this rich information for improved performance (e.g., Jedermann et al., 2009, Laniel et al., 2011 and Piramuthu et al., 2012). To our knowledge, there is a lack of the existing literature that cover this topic. We attempt to address this gap in current literature by studying inventory management of perishable items using item-level information generated through semi-passive RFID tags with appropriate (e.g., temperature) sensors. Specifically, we extend the model presented in Bai and Kendall (2008) to include RFID-generated item-level information and consequently model the facings based directly on the quality profile of items on the shelf. We solve the resulting problem using Genetic Algorithm-based Evolutionary Solver. Evolutionary Solver is Frontline Systems' implementation of Genetic Algorithm in their Premium Solver for Education (solver.com) product that runs on Microsoft Excel Spreadsheets.
The contribution of this paper is three-fold: (1) we model the number of facings at a finer (i.e., at the item level) level of granularity based exactly on the instantaneous quality of each individual items on the shelf, (2) we consider the possibility for the retailer to display items based purely on their remaining shelf-life, and (3) we show that the dynamics when item-level information are available is disparate from the case where only class-level information is available. The way we model the number of facings based purely on each of the item's instantaneous quality is significant for perishables since, for example, any number of perishables with quality below consumable-level will not generate demand, i.e., number of facings has to take the quality of the items into account and none of the existing literature does that, to our knowledge.
The rest of the paper is organized as follows: we provide a brief background on the domain of interest to this study in the next section. We then present our model and analysis in the following section. We conclude with a very brief discussion in the concluding section.
