Business Challenge
Retail suppliers struggle to balance inventory levels for perishable goods, needing precise sales forecasts to avoid overstocking or stock-outs.
Data Analysis Approach
● Data Source: 714-day sales data from major retail stores, provided by Nuqleous.
● Forecasting Methods:
○ Direct Forecasting of total weekly sales.
○ Channel-Specific Aggregation across different sales channels.
Key Forecasting Models
Employed models like Naive, Seasonal Naive, ETS, TSLM, and ARIMA, tailored to specific data segments.
Core Findings
● Cost Efficiency: Channel-specific aggregation methods generally proved more cost-effective than
direct forecasting.
● Performance Note: Similar results for both methods when using regression, Naive, or Seasonal
Naive models.
● Exception Case: SKU 1538336 showcased better performance in the General Level during
validation.
Recommendations
● Data Duration: Minimum two-year data span for robust forecasting.
● Benchmarking: Essential to compare new methods against existing standards.
Conclusion
Channel-specific forecasting tends to yield better cost outcomes, except when certain models like regression or Naive are used, underscoring its importance in effective inventory management and ROI enhancement.