Skip to main content

AI Forecasting In Depth

How FLYR's AI forecast is generated

Michael Sandéen avatar
Written by Michael Sandéen
Updated over a week ago

Overview

Our AI forecasting system estimates daily revenues by property up to two years in advance. Rather than predicting revenue directly, it models three key components:

  • Extra arrivals – how many guest are expected to show up

  • Length of stay – how long they’ll stay

  • Average Daily Rate (ADR) – how much they’re likely to pay per night

These elements are combined with existing bookings to create the final revenue forecast.


Clustering

To improve accuracy, we group similar rate plans using a clustering algorithm. Each group is forecasted separately for arrivals and length of stay to reflect its unique booking patterns.


Extra Arrivals Forecast

The arrivals forecast is the most challenging and refined part of our system. It estimates how many additional guests will arrive each day and adapts depending on how far into the future we’re forecasting.


Standard vs. Special Days

We treat holidays and notable dates differently from regular days, using separate models tailored to each.


Forecasting Approach

  • Long-term forecasts rely mostly on historical patterns from the same time last year.

  • Mid-term forecasts adjust last year’s data based on how current performance compares.

  • Short-term forecasts use a machine learning model trained on patterns from across our customer base, combining recent booking trends with historical data.

For holidays and events, we match past dates with similar characteristics and use their performance to guide the forecast.


Length of Stay Forecast

This forecast predicts how long arriving guests are likely to stay. It considers both the day of the week and how far in advance bookings are made. Simpler models are used for longer-term forecasts, while short-term ones incorporate more factors for better precision.


ADR Forecast

The ADR forecast predicts average nightly rates at the inventory type level and adjusts it across rate plans for consistency.

  • Short-term rates are based on recent trends, current bookings, and historical data.

  • Long-term rates are influenced more by seasonal and day-of-week patterns.

  • Special days use past event pricing or fall back to current on-the-books rates when needed.


By layering historical trends with real-time data, our forecasting system helps properties make smarter, more confident decisions about pricing and planning.

For more on the Forecasting topic, we can recommend this article.

Did this answer your question?