Forecasts

Predict KPI Trends Over the Next 12 Weeks

Forecasting helps you anticipate the behavior of key performance indicators (KPIs) for the upcoming 12 weeks, enabling better planning and decision-making.

The forecast interface provides:

• Predicted KPI Values: See the expected value for each KPI along with its denominator (e.g., the number of users, if applicable).
• Confidence Intervals: Each forecast includes a 95% confidence interval, giving you a range within which the KPI is likely to fall.

How Forecasts Are Generated

Loops leverages the widely respected open-source forecasting tool Prophet to deliver reliable and interpretable predictions. Prophet uses an additive model to decompose time-series data into its components, capturing:

• Yearly Seasonality: Recurring patterns influenced by holidays, fiscal quarters, or other annual events.
• Weekly Seasonality: Behavioral changes across weekdays and weekends.
• Trend Dynamics: Long-term growth or decline in the KPI.

To generate the forecast Loops' uses all available history of the KPI (by default this is 6 months before when the KPI or funnel was created for weekly and daily KPIs and 12 for monthly KPIs).

To account for yearly seasonality effects, at least 2 years of historical data is required for optimal accuracy.

Why This Method Was Chosen

The Prophet method was selected after extensive research and rigorous testing on thousands of historical signals from dozens of products. Its reliability and adaptability make it ideal for capturing seasonality and long-term trends across diverse industries and data behaviors.

• Fourier Series: The method decomposes seasonal patterns into Fourier series, allowing it to model complex periodicities efficiently.
• Accuracy: With sufficient historical data, the model achieves a median absolute error of about 10% for the furthest (12th) week of forecast, making it a trusted choice for strategic planning.

Important Notes

Loops employs different models for different purposes:

• Forecasting: Focused on long-term trends using Prophet for robust predictions. The forecasts were optimized based on historical data where the future is known.
• Anomaly Detection: Loops employs a proprietary algorithm specifically designed for next-point predictions, offering superior accuracy for short-term monitoring. This model was trained on thousands of (professional) human generated tags of data anomalies.