Booking volume and revenue forecasts at the property, route, or market level over rolling 30-90 day horizons. Trained on your historical booking data, search query volumes, and external event signals. Outputs uncertainty-bounded forecasts that feed inventory allocation, group pricing decisions, and revenue budget planning. For hotels, forecasts at the room-type level. For airlines, forecasts by origin-destination and cabin class. Replaces spreadsheet-based occupancy and load factor projections with model-driven forward-looking estimates.
The forecasting model uses LightGBM gradient boosting as the primary estimator for its handling of irregular seasonal patterns, missing data, and mixed-type feature sets common in travel booking data. Inputs include historical booking pace curves (pickup by lead time window for the equivalent period in prior years), search query volumes from your platform (search-to-book conversion rate as a leading demand indicator), competitor pricing history, local event calendars sourced from Eventbrite or PredictHQ, and macroeconomic signals where relevant for longer-horizon forecasts. For NLP-based itinerary demand signals, extracting travel intent from unstructured booking notes, group inquiry emails, or tour operator correspondence, spaCy NER and transformer-based classification models parse origin, destination, travel party size, and date range from free text at ingestion. Forecast output includes point estimate plus confidence interval bands so your revenue management team can distinguish high-certainty near-term forecasts from wider-ranging long-horizon projections. Forecast accuracy is tracked against actuals using MAPE and symmetric MAPE by forecast horizon, with model retraining triggered automatically when drift exceeds calibrated thresholds.
