RevPAR, ADR and occupancy: how to forecast hotel KPIs with AI

What each KPI means

• Occupancy: rooms sold over rooms available.
• ADR: revenue per room sold.
• RevPAR: occupancy × ADR. The king KPI.

Why predict all three together

If you forecast 90% occupancy but ADR drops 20%, RevPAR falls. The three are correlated and serious models predict them as a system.

Models that work in 2026

• Gradient boosting (XGBoost, LightGBM): the workhorse.
• Time models (Prophet, NeuralProphet): capture seasonality.
• Hybrid models with event embeddings: state of the art.
• LLMs as explanation layer, not pure prediction.

Variables feeding the forecast

1. Booking and cancellation history.
2. Local events calendar.
3. Historical and forecast weather.
4. Google search trends.
5. Current pickup vs same date prior year.
6. Compset (estimated competitor prices and occupancy).
7. FX rates for foreign source markets.

How to measure accuracy

• 30-day MAPE.
• Acceptable target: <8% occupancy, <5% ADR.
• Beat human forecast.
• Stable error across seasons.

How to integrate into operations

Forecast is useless without delivery. AI must hand it over each morning in plain language: "This week closes 4 points below forecast on weak midweek demand. Recommend launching Sat-Sun 2x1 promo."

Want a real 90-day forecast for your hotel? Request a demo.

Frequently asked questions