A percentile marks a chosen position in a distribution: a 95th percentile represents a forecast target associated with a 95% service level. In demand forecasting, choosing the right percentile is a business decision because higher coverage can reduce out-of-stocks while affecting inventory and forecast evaluation. section 5.3.1
Demand-forecasting baseline
A grocery retailer can calculate 1.5th, 25th, 50th, 75th, 95th, and 99th percentiles from a yearly window and apply them to a baseline forecast. A linear model can also use quantile loss with sales lags, rolling aggregates, price, revenue, or customer counts. The selected percentiles can be weighted by SKU price and adjusted when underforecasting or overforecasting has different costs. section 8.6.1 section 5.3.1
Percentiles connect forecast quality to operations rather than treating every error equally. Quantile forecasts can feed stock-control strategies, while quantile metrics can be reported with confidence intervals and compared across products or groups. section 5.3.1
