A quantile forecast targets a chosen point in a predictive distribution rather than its average. For a retailer, the useful choice may be the quantile matching a service objective, such as 95% availability. section 5.3.1
Use quantile forecasts when decisions must balance excess stock against stockouts:
- Feed selected quantiles into stock-control strategies, then judge forecast quality by total stock position and out-of-stocks over time. section 5.3.1
- Choose loss functions or scoring rules that elicit the target quantile, keeping the loss aligned with the business metric. section 5.3.1
- Add SKU-price weights or underforecasting and overforecasting penalties when particular items or error directions matter more. section 5.3.1
Service-level forecasting
A retailer can evaluate quantiles such as 1.5, 25, 50, 75, 95, and 99, reporting point estimates with 95% confidence intervals and optionally converting results to percentage errors. section 5.3.1
