Demand Forecasting

Overview

Use this skill to turn sales, inventory, branch, and operational signals into demand forecasts and replenishment recommendations. It is especially relevant when fixing SQL joins that duplicate products, deriving days until stockout, or documenting demand-driven planning assumptions.

Workflow

1. Define the reporting grain first: usually one row per product per shop, branch, outlet, or warehouse for the forecast horizon.
2. Aggregate sales and stock movements before joining product, branch, and stock-balance tables. Do not join raw sales lines directly to item master or stock balances when the output expects one product row.
3. Exclude or separately flag voided sales, returns, internal transfers, stockout days, and one-off events that would distort demand.
4. Normalize demand to a daily rate. Use 7, 30, and 90 day windows when available, and explain which window drives the forecast.
5. Derive days until stockout as current_stock / daily_demand. If demand is zero, report "no active demand" rather than hiding the value as an unexplained N/A.
6. Calculate forecast consumption as daily_demand * horizon_days.
7. Calculate reorder point as daily_demand * lead_time_days + safety_stock.
8. Calculate suggested order as max(0, forecast_consumption + safety_stock - current_stock - inbound_qty).
9. Backtest against historical periods using WAPE/MAPE, bias, and missed-stockout counts.

Join Guardrails