Machine Learning

Definition

Models that learn patterns from historical outcomes instead of following hand-written formulas. In hotel revenue work this is the layer behind demand forecasting, anomaly detection and rate recommendation: it generalises from what your bookings did before to what they are likely to do next.

What it tells you

ML is strong exactly where human attention is weak: it watches every date, segment and day type with the same care, every day, and surfaces the small, cumulative drifts that look like noise date by date. Its systematic blind spot is the unprecedented — a new competitor, a lost flight route, a regime change — where the past it learned from no longer applies.

How to track it

Judge a model by its track record, not its confidence: monthly forecast-accuracy reviews (MAPE by horizon), and a log of when you overrode it and whether the override beat the model. If overrides consistently win in a period, the world has probably shifted faster than the training data.

Where it fits

The third step of the pricing evolution — after rule-based logic and elasticity models — and the statistical base under hybrid forecasting. The free Academy covers it in depth: Pricing Engine — ML-based rate recommendation.