An inventory-aware and revenue-based itemset placement framework for retail stores

Highlights

• We introduced the notion of inventory in retail itemset placement.

• Inventory-aware itemset placement can significantly improve retailer revenue.

• We have proposed an inventory-aware itemset retrieval and placement framework.

• We conducted experiments with real datasets to demonstrate revenue improvement.

Abstract

Retailer revenue is significantly impacted by item placement. Given the prevalence and popularity of medium-to-large-sized retail stores, several research efforts have been made towards facilitating item/itemset placement for improving retailer revenue. However, they fail to consider the issue of inventory of the items w.r.t. itemset placement. Notably, the inventory of a given item refers to the number of instances of that item that are available to the retailer for sales purposes. Moreover, efficient retrieval and placement of top-revenue itemsets in the retail store slots cannot be performed by existing approaches. Our key contributions are summarized as follows. First, we introduce the notion of inventory in retail itemset placement. Second, we propose an inventory-aware indexing scheme, designated as IRIS, for efficiently retrieving high-revenue itemsets. Moreover, we propose the IRPS inventory-aware itemset placement scheme, which exploits the IRIS indexing scheme, for facilitating improved retailer revenue. Third, we conduct a performance study with two real datasets to demonstrate the effectiveness of our proposed itemset indexing and placement schemes in improving retailer revenue.

Introduction

During the past few decades, large retail stores have been becoming increasingly popular (lar, 2019). Examples include Walmart Supercenters, Dubai Mall (Dubai) and Macy’s Department Store in New York City. Such large stores can have retail floor space of more than a million square feet. It is a well-established fact that retailer revenue is significantly impacted by item placement in such retail stores (Chen et al., 2006, Chen and Lin, 2007, Hansen and Heinsbroek, 1979).

Typically, items are placed in the retail store slots. The goal is to place a large number of items such that retailer revenue is improved. Notably, retail store slots can either be premium (i.e., high-visibility/accessibility slots) or non-premium (i.e., low-visibility/accessibility slots). In practice, items placed in the premium slots have a considerably better probability of sale than items that are placed in non-premium slots due to better visibility/accessibility. For example, items placed in slots near to the checkout counters or items placed in slots nearby the eye level of users tend to be more visible/accessible; hence, they are likely to sell more quickly. Hence, item placement in premium slots significantly impacts retailer revenue (Ahn, 2012, Chaudhary et al., 2019, Chaudhary et al., 2020).

Incidentally, in the context of retail stores, customers often prefer to buy itemsets as opposed to individual items. Examples of itemsets could be {bread, jam, milk}, {Pepsi, biscuits, peanuts} and so on. From the customers’ perspective, one-stop shopping adds value in terms of convenience. Hence, there are efforts in the literature for improving the retailer revenue by placing itemsets (Ahn, 2012, Chaudhary et al., 2017, Chaudhary et al., 2018, Chaudhary et al., 2019, Chaudhary et al., 2020), which are extracted from the user purchase transactions. By extracting and carrying out placement by exploiting the knowledge of itemsets, it is possible to improve the revenue. Consequently, this work considers the placement of itemsets only for the premium slots.

Existing works focus on stock-out management (Breugelmans et al., 2006, Corsten and Gruen, 2003), inventory management (Caro & Gallien, 2010), supply chain management (Bhattacharjee & Ramesh, 2000), association rule mining (Agrawal and Srikant, 1994, Han et al., 2000, Pasquier et al., 1999), identification of high-utility itemsets (Chan et al., 2003, Fournier-Viger, Lin, et al., 2016, Fournier-Viger, Wu, and Tseng, 2014, Fournier-Viger, Wu, Zida, and Tseng, 2014, Fournier-Viger, Zida, et al., 2016, Hong et al., 2011, Jaysawal and Huang, 2019, Krishnamoorthy, 2019, Lin et al., 2019, Liu et al., 2005, Liu and Qu, 2012, Liu et al., 2015, Mai et al., 2020, Nguyen et al., 2019, Truong et al., 2019, Tseng et al., 2015, Tseng et al., 2010, Zida et al., 2015), incremental utility mining (Kim et al., 2020, Lee et al., 2018, Nam et al., 2020, Vo et al., 2020, Wu et al., 2020, Yun et al., 2019) and retail itemset placement (Ahn, 2012, Chaudhary et al., 2017, Chaudhary et al., 2018, Chaudhary et al., 2019, Chaudhary et al., 2020). Notably, none of the existing works address the issue of inventory-aware retail itemset placement. Hence, we address the problem of inventory-aware retail itemset placement for improving retailer revenue.

Existing utility mining approaches suffer from two important drawbacks w.r.t. the problem of itemset placement in retail stores. First, they do not consider the issue of inventory. Here, the inventory of a given item is the number of instances of that item that are available to the retailer for sales purposes i.e., these instances of that item can be placed in the retail store slots. For example, if the retailer has 1000 bottles of Pepsi available for sale, the inventory value for Pepsi would be 1000. In other words, existing works implicitly assume that the inventory of each item is essentially infinite, and this assumption is not true in practice (as we shall describe shortly in a subsequent paragraph).

Observe that since existing works do not perform itemset placement in an inventory-aware manner, they could identify itemsets (for placement), where one or more of the items in those itemsets lack adequate inventory. For example, consider an itemset {A, B, C}, where the individual inventory values of A, B and C are 100, 150 and 200 respectively. Observe that this itemset could be sold at most 100 times after which the inventory of A would become zero, thereby precluding the possibility of any further sales of this itemset. Since existing works do not consider this important issue, they could erroneously try to repeat the placement of this itemset more than 100 times (since they assume unlimited inventory), thereby in effect failing to improve the revenue of the retailer. In essence, itemset placement has to take the inventory values of the items into account in order to be effective towards improving retailer revenue.

Second, efficient retrieval and placement of top-revenue itemsets in the premium slots cannot be performed by existing approaches. Incidentally, in our previous works on utility-based retail itemset placement, we have focused on indexing techniques for efficiently determining high-revenue itemsets (Chaudhary et al., 2017, Chaudhary et al., 2020), itemset placement with diversification (Chaudhary et al., 2018), itemset placement in slots with varied levels of premiumness (Chaudhary et al., 2019) and urgency-aware itemset placement (Mittal et al., 2021). However, like other existing utility mining approaches, our previous works also have not addressed the issue of inventory.

In practice, the amount of inventory (of any item), which is available to a retailer, is limited and fixed i.e., a retailer cannot have access to infinite inventory for any item. This happens due to several practical reasons such as the cost of items, storage space requirements for items, insurance costs (e.g., pertaining to fire hazards) and so on. Furthermore, inventory cannot be replenished by the retailer on an immediate or on-demand basis. This is because items typically come to the retailer through the supply chain. Supply chain contracts in the retail industry generally need to be in place at least several weeks before the items physically reach the retail store. Supply chain contracts in the retail industry generally require time to be spent on contract negotiations (in terms of item prices and quantities, refund policies, and terms and conditions). Once such supply chain contracts have been negotiated, transportation and logistics of items also require time. Hence, in consonance with real-world scenarios, we assume that the amount of inventory (for any item), is limited and fixed, and retailers cannot replenish inventory on an on-demand basis.

Incidentally, in practice, retailers perform the placement of items (or itemsets) in the slots of the retail stores only periodically as opposed to replenishing slots with items as soon as any item gets sold. The periodicity essentially depends upon the business strategy of the retailer. For example, a given retailer could decide to perform itemset placement once every 24 h during a fixed time of day e.g., at 7 am before the retail store opens. Hence, in consonance with real-world retail scenarios, this work also assumes that the placement of itemsets occurs periodically i.e., we assume that itemsets are placed in the retail store slots by the retailer for a given period of time. Within this period of time, the slots, which become empty due to items being purchased by customers, are not replenished with any items.

This work addresses the problem of inventory-aware retail itemset placement for improving retailer revenue. Given a finite set of items, we examine the history of user purchase transactions, from which we extract itemsets. The problem is to (a) identify high-revenue itemsets with consideration of the respective inventory values of these itemsets and (b) place these itemsets in the required number of available (retail store) premium slots to improve retailer revenue.

In this regard, we propose an inventory-aware indexing scheme, which we designate as Inventory-aware top-Revenue itemset Indexing Scheme (IRIS), for efficiently retrieving high-revenue itemsets. IRIS stores itemsets at each of its levels in an inventory-aware manner. Each level of the IRIS index stores top-$\lambda$ itemsets (based on revenue) of a specific size e.g., level 3 of IRIS stores itemsets of size 3. Observe that storing only the top-$\lambda$ itemsets at each level instead of storing all of the possible candidate itemsets considerably reduces candidate itemset generation overhead. Furthermore, we propose an itemset placement scheme, which we designate as Inventory-aware and Revenue-based Placement Scheme (IRPS), for facilitating improved retailer revenue. IRPS exploits the IRIS index in order to identify the high-revenue itemsets for placement purposes. IRPS considers both expected revenue as well as inventory of the itemsets towards performing itemset placement.

Our key contributions are summarized below:

• We introduce the notion of inventory in retail itemset placement.

• We propose an inventory-aware indexing scheme, designated as IRIS, for efficiently retrieving high-revenue itemsets. Moreover, we propose the IRPS inventory-aware itemset placement scheme, which exploits the IRIS indexing scheme, for facilitating improved retailer revenue.

• We conduct a performance study with two real datasets to demonstrate the effectiveness of our proposed itemset indexing and placement schemes in improving retailer revenue.

This is the first work to address inventory-aware placement of itemsets in retail stores. The organization of the paper is as follows. Sections 2 Related work, 3 Proposed framework of the problem discuss related works and proposed problem framework respectively. Sections 4 Inventory-aware top-revenue itemset indexing scheme (IRIS), 5 IRPS: Our proposed inventory-aware and revenue-based itemset placement scheme present our proposed indexing and placement schemes respectively. In Section 6, we report the performance study. Section 7 discusses the managerial implications and limitations of our proposed itemset placement framework. Finally, we conclude in Section 8 with directions for future work.