Using Fatigue Models the Right Way

Mathematical modelling is often misunderstood because people expect it to predict individual events with precision. But models are usually far better at describing patterns across many events than forecasting a single outcome.

Consider throwing two dice once. Predicting the exact sum is essentially impossible; any result from 2 to 12 could appear. Yet if we throw the dice 1,000 times, the average sum becomes highly predictable. The mathematics tells us the average will settle very close to 7. The larger the sample, the more reliable the prediction becomes.

Fatigue risk modelling works in a similar way. These models are valuable when designing pairings, constructing rosters, and assessing the overall operating environment across many flights and crews. In that context, they help identify broad patterns of elevated fatigue exposure and support better system-level decisions.

Problems arise when a model designed for population-level insight is used as a near-real-time “go/no-go” gauge for a single flight close to departure. At the individual-flight level, uncertainty is much higher and operational reality is influenced by many factors the model cannot fully capture. Treating the model output as a definitive inspection tool can create a false sense of precision. And elevate costs.

A more effective approach is to use fatigue modelling as part of process control: shaping schedules and operational practices to reduce fatigue risk before it occurs, rather than relying on late-stage inspection and re-work. This document provides a useful comparison between these two approaches and explains why process control generally delivers better outcomes.