Definition
Degree to which the overbooking model's predicted no-show rate matches the actual no-show rate at the flight level across route types and booking windows: measuring whether the model remains calibrated to current traveler behavior or has drifted from the behavioral distribution it was trained on.
The Measurement Gap
Overbooking performance is traditionally measured by its consequences: denied boarding rate, involuntary bumping rate, and voluntary bump cost per seat. These are outcome metrics: they reveal that overbooking was wrong only after passengers are already denied boarding. They cannot detect the direction or magnitude of model miscalibration before consequences materialize. Denied boarding reports aggregate across all flights and hide the structure of the problem: a model that over-predicts no-shows on business routes and under-predicts on leisure routes will show an acceptable average denied boarding rate while creating systematic problems in specific segments that are invisible to the aggregate metric. RASM and yield incorporate overbooking revenue effects but cannot distinguish the AI model's calibration from pricing decisions. No standard airline metric reports whether the model's predicted no-show rate for flights in a specific confidence band actually matches reality in that band: the definition of calibration. This means calibration drift is invisible until consequences spike. OCS detects drift before denied boardings increase, and identifies which route types or booking windows are driving the problem so that targeted recalibration can be applied before the denial rate signals trouble.
Decision Supported:
Is the overbooking model calibrated to current traveler behavior, or has it drifted in ways that create systematic denied boarding or revenue leakage risk?
Formula
Plain language: Group flights by the model's predicted no-show rate band. Within each band, compare the average predicted no-show rate to the actual no-show rate. Perfect calibration means predicted equals actual in every band. Score is the average absolute calibration error subtracted from 1.0.
Notation:
OCS = 1 − (1/B) × Σ |Predicted_noshow_rate_b − Actual_noshow_rate_b|
Variables:
| Variable | Definition | Unit |
|---|---|---|
| B | Number of no-show rate bands (recommended: 5 bands from 0% to 25%+) | Count |
| Predicted_noshow_rate_b | Average predicted no-show rate for all flights assigned to band b by the overbooking model | Percentage |
| Actual_noshow_rate_b | Actual no-show rate for flights in band b, measured at departure | Percentage |
Also track directional bias separately: is the model systematically over-predicting no-shows (leading to over-overbooking and higher denied boarding risk) or under-predicting (leaving revenue on the table through under-overbooking)?
Proposed formula, pending validation pilots.