DARK STORE PERFORMANCE SEGMENTATION IN INDIAN QUICK COMMERCE: A K-MEANS CLUSTERING AND HYPOTHESIS-TESTING APPROACH

Abstract

Quick commerce, or Q-commerce, has become one of the most transformative trends in the retail sector in India, offering the ability to deliver groceries and everyday items within 10 to 30 minutes through a network of nearby dark stores. Although this model has grown quickly, operators often lack a structured approach to understand the varying levels of performance among these dark stores. This study fills that gap by using unsupervised machine learning and statistical methods to analyze sales data from 60 dark stores in a large Indian city. The data covers 13 categories of fast-moving consumer goods and groceries, excluding electronics to avoid bias in the average order value. The process includes creating features such as average order value, discount percentage, and product mix from the transaction data. K-Means clustering is applied to the standardized and log-transformed data to group the dark stores into four distinct performance categories. Principal Component Analysis is used to visualize these clusters in two dimensions. The results are checked using statistical techniques like one-way ANOVA, Welch’s t-test, and Kruskal-Wallis test to test six hypotheses. Five out of the six hypotheses are confirmed at a 5% significance level, showing that the different clusters perform significantly differently in terms of total sales, number of orders, average order value, and discount patterns. The four types of stores identified are Moderate Performers, Anomalous/Inactive Stores, Premium/Sparse Stores, and High-Volume Workhorses. These insights help in making informed decisions about inventory management, pricing, and marketing efforts across the dark store network. This research represents one of the first detailed analyses of dark store performance in the Indian quick commerce sector, offering both a practical analytical method and strategic advice for operators.