Inventory Optimization: 18% Sales Growth Through Statistical Demand Forecasting
Overview
A multi-location shoe retailer was experiencing declining sales due to chronic stockouts in popular sizes, while excess inventory sat unsold. I developed a statistical inventory optimization system that increased sales by 18% in the first year, reversing the company's decline.
Details
Inventory is an essential component in the retail industry. Maintaining the right stock levels in a dynamic market can be challenging. In this project, I employed multiple methods to analyze the data of a female shoe store, enabling management to determine the optimal levels of stock based on style and size.
Problem Statement
The shoe store in question consistently found itself in a position where it was losing sales because it did not have the right sizes or styles in stock. Additionally, a significant opportunity was being missed due to certain shoes sitting in stock for an extended period, which were ultimately sold at a heavily discounted price.
The Data
The data provided included sales from multiple store locations spanning several years. After our EDA and initial analysis, we determined there was a concept drift.
Concept Drift: Due to shifting fashion trends and customer preferences over time, patterns from three years ago or more no longer apply to our analysis. We established a 2-year lookback window to ensure the training data reflected current market dynamics.
Statistical Approach to Inventory Optimization
To address the inventory management, I used inferential statistics. It allowed me to draw conclusions and make predictions by looking at only a sample population. Analyzing shoe sizes, styles, and locations, I identified sales patterns based on products and locations.
I employed stratified sampling by location and style, calculating 95% confidence intervals for each SKU’s demand distribution. This approach revealed that Store A needed 25% more athletic styles in sizes 7-8, while Store B’s demand skewed toward formal footwear in larger sizes.
Additionally, this statistical method provided two critical bounds:
Lower Bound: The minimum inventory level expected to meet demand.
Upper Bound: The maximum inventory level that should be maintained
The confidence interval approach accounts for demand variability and data uncertainty. By using the proper bound as the recommended inventory level, I ensured that the business would have sufficient stock to meet customer demand 95% of the time, balancing the costs of overstocking against the risk of stockout.
The final recommendation rounded the upper bound to provide practical, actionable inventory targets that align with the client’s operational constraints.
Results
Following the full implementation of my analysis, sales increased by 18% in the first year, reversing the company’s sales decline and returning it to growth.
The most important element was to ensure the store had the right inventory for its clientele. Each store had to be treated differently as they had different demands, and the styles and sizes varied drastically.