Status : Verified
|Personal Name||De La Vega, Sandi D.|
|Resource Title||A Store-based grocery assortment planning model incorporating variety and a substitution mechanism|
|Date Issued||26 July 2019|
|Abstract||Assortment Planning (AP) under Retail Category Management (RCM) practice continues to be a challenge to store-based grocery retailers due to the need to offer a large variety of product while maintaining profitability. By balancing the trade-off between product proliferation and store’s limited carrying capacity, a delisting strategy called internal SKU rationalization (ISR) is needed to remove unprofitable SKUs from the store’s assortment. By doing so, the store-based grocery retailers can gain short-term profitability because of the reduction in carrying cost reduction and the additional profit gained from transferable demand of delisted SKUs defined by an assortment-based substitution mechanism, where the remaining listed SKU can substitute a portion of the delisted SKU's demand in the assortment. Mathematically, this mechanism is defined using sales volume based on historical data. However, by focusing only on the sales volume and profit alone, the danger of delisting too much SKUs can lead to a significant reduction in store sales and retail patronage when the assortment has a wide variety of products and has low-frequently purchased subcategories. Therefore, there is a need to develop a Grocery RCM Assortment Planning model focusing on Internal SKU Rationalization (GRAP-ISR) that incorporates variety and an assortment-based substitution mechanism to delist unprofitable SKUs and maintain profitability.
Using the Category Level Model Approach (CLMA), Gower Distance Measure, Subcategorization and Similarity Effects Concept, a Grocery RCM Assortment Planning Optimization Model (GRAPOM) is developed using monthly aggregated sales data and SKU attributes as input data for the GRAP-ISR framework developed in this study. The GRAPOM and Fadilog̃lu et al.  (FKP) model is implemented using a Python-based optimization package called PYOMO and open-source AMPL solver called CBC solver. In the absence of appropriate attribute-based metrics to evaluate the two AP models, the results are compared in terms of monthly profit, total no. of SKU listed (TSL), SKU retained ratio (SRR), sales volume retained ratios (VoRR and NVoRR) and newly defined metrics called unweighted average accuracy (UAA) and weighted average accuracy (WAA) in this study. Similar to how accuracy is applied to evaluate supervised learning algorithms, the WAA metric is an accuracy of the AP model compared to the desired assortment of the retailer with weights assigned per CLMA subcategory based on total sales volume, where low-frequently purchased subcategories has higher weights, while the UAA metric assumes equal weights per CLMA subcategory. Among all the metrics, the key performance indicators of the AP model are the TSL, SRR, NVoRR, and WAA.
As an attribute-based extension to FKP model, GRAPOM also utilizes the minimum conservation ratio (MCR) which is the minimum ratio of remaining listed sales volume compared to the initial listed sales volume for each CLMA subcategory. The advantage of the GRAPOM is the use of subcategory MCR and CLMA subcategory MCR; both defined in terms of non-transferable demand sales volume. Using this flexibility of choosing different MCR, the GRAPOM achieves significantly higher values of WAA metric than the FKP model.
For the case study of one store, the GRAPOM reached approximately 79.74% to 87.43% in WAA; in contrast, the FKP model reached approximately 77.17% to 85. 95% in WAA. Also, the GRAPOM achieves significantly higher values of TSL, SRR, and NVoRR than the FKP model at the category level performance. GRAPOM performs significantly better in delisting SKUs conservatively and accurately than the FKP model, especially to the low-frequently purchased subcategories.
|Degree Course||Master of Science in Industrial Engineering|
|Keyword||Assortment Planning; Attributes; Category Level Model Approach; CBC; Gower Distance; PYOMO; Similarity Effects; Substitution Mechanism; Variety|