Seasons in PMI are automatically defined periods that group together historical dates with similar booking behavior or activity levels. They help PMI make accurate forecasts by comparing upcoming dates with past periods that behaved in a comparable way.

How Are Seasons Created?

• Generated by Machine Learning (ML): PMI automatically analyzes historical data using ML to determine which dates belong in the same “season” — no manual setup is needed.
• Based on Similar Patterns: Seasons are not tied to calendar months. For example, early December (high activity) and late December (low activity) will belong to separate seasons because their patterns differ.
• Flexible Length: The length of a season varies depending on data patterns and department activity. PMI adjusts season length dynamically — there’s no fixed rule like “2–3 weeks.”
• Historical Depth: PMI may look back up to 5 years, but the exact timeframe depends on the department and the availability of relevant data. Older data is used only if it’s still considered relevant.

What About Outliers?

Some dates behave abnormally — for example, due to one-off events like concerts, renovations, and some dates have inaccurate data due to for instance revenue adjustments from a previous date.

• Automatically detected: Most outliers are flagged by PMI’s system.
• Manually added: If needed, users with the right access can manually label additional outlier periods via Live Forecast > Tools > Seasons.
• Outliers are excluded from forecast calculations to avoid skewing the results.

Additional Notes

• There is no longer an option to manually define seasons per department — the ML model handles this automatically for each department based on its own data.
• The idea of “each month as a season” or manually grouping seasons is no longer supported. PMI’s model performs better when it handles this automatically.