WHAT WE PROVIDED
Steps
Actions
Output/refinement
Data gathering and categorization
The team collected essential data, including scanning information, prior shrink records, and store attributes. Split data into three subsets based on shrinkage percentages for focused model training. After cleaning the data to remove nulls and other anomalies, a simple linear regression was trained on each subset, assessing model quality using metrics like p-values and directional shrinkage numbers per their business implications. Refined division-level correlations were computed between actual and predicted shrinkage, leading to adjustments in predicted shrinkage percentages. When division correlations surpassed 50%, corresponding adjustments were made to trigger an alarm. Caps were also implemented- to eliminate pessimistic or overly high seasonal predictions, resulting in more accurate and actionable final shrinkage predictions in both percentage and dollar terms. This helps in informing strategic decisions and resource allocation for efficient operational responses.
Data preparation led to subsets for specialized model training, streamlining predictions based on shrinkage categories. Model quality evaluations enhanced prediction accuracy and refined models for reliable shrinkage forecasts.
Data cleaning, model training and evaluation
Prediction refinement and adjustment
Adjusted predictions resulted in more precise alarms & refined shrinkage estimates, thus fine-tuning predictions to make them more reliable & ensuring actionable & accurate insights.
Prediction caps and operational implementation
Implementing translated refined predictions led to actionable insights for better planning and resource allocation.
Made with FlippingBook - PDF hosting