Supply Chain Demand Forecasting and Inventory Optimization

Problem Context

When operations teams have date-level sales data but need answers like next-month demand, reorder timing, and high-risk SKUs, ad-hoc spreadsheets can easily lead to stockouts or excess inventory. This skill targets CSV/Excel sales records and turns demand forecasting into a repeatable analysis workflow.

How It Works

Input files usually include date, sku, and sales, with optional fields such as price, promotion, stock, and lead_time. The workflow starts with data quality checks for missing values, time-series continuity, outliers, and seasonality, then produces cleaned_data.csv. Forecasting can select methods such as Prophet, XGBoost, or LightGBM, with Prophet as the default when data volume is uncertain and seasonal or holiday effects matter. Results include confidence intervals in forecast_results.csv and an interactive chart. If inventory parameters are provided, it can also estimate safety stock, reorder point ROP, and EOQ, then export replenishment recommendations. The final HTML report summarizes historical trends, decomposition, model comparison, MAPE, MAE, RMSE, and high-risk SKUs.

Boundaries and Caveats

It works best with at least 12 continuous time points and a clear target such as a SKU, category, or total demand. If missing data is heavy, outliers may reflect real business events, or MAPE remains high, the outputs should be reviewed with promotions, holidays, production stops, or channel changes in mind rather than treated as final decisions.