Problem Statement:
A popular retailer with over 3000 physical stores in the United States wanted to increase revenue from each of their stores by improving merchandise effectiveness in their physical stores.
Solution:
We developed a Merchandise Mix Optimizer using machine learning and data analysis techniques. The model determined the catchment area for each store to identify the opportunity to sell products offline and projected the sales potential of new products using offline and online sales data of similar existing products. The ML model continuously optimized the merchandise mix based on the correlation between product-level demand forecasts and actual sales at individual stores.
The model used census data, population density, geographical factors, and driving distance to determine the catchment area for each store. For each SKU, product attributes and buyers’ demographic profiles were factored in to predict offline sales. A genetic algorithm was used to find the optimum weights to be assigned to each product attribute based on historical data, and a sales forecasting model was developed for new products based on the similarity to existing products. The technical solution was developed with Python, SQL Server, and machine learning techniques including Random Forests, Spark, Genetic Algorithm, and Neural Network.
Benefits:
Testing of this model for small geography showed great potential to optimize the merchandise mix in the store, leading to a significant increase in sales. By identifying what products should be introduced, stored, or discontinued at each store based on forecasted revenue and profits, the retailer was able to improve merchandise effectiveness and increase revenue from each of their stores.
