Forecasting is a key part of planning.
Many organizations use historical data, spreadsheets, and business assumptions to estimate future results. Workday Adaptive Planning includes Predictive Forecaster to help generate forecasted data using machine learning algorithms.
In this article, we look at what Predictive Forecaster is, how it works, and the requirements to consider before using it.
What Is Predictive Forecaster?
Predictive Forecaster is an Intelligent Planning capability within Workday Adaptive Planning.
It uses machine learning algorithms to generate forecasted data based on historical reference data.
The forecasted values are written into a selected plan version and can be used as a starting point for planning and budgeting activities.
According to Workday documentation, Predictive Forecaster is designed to seed plan versions with machine learning generated data that budget managers can review and adjust.
How Predictive Forecaster Works
The process starts with historical data.
Users select:
- A standard or cube sheet
- A forecast version
- Historical actuals data
- A forecast period
- A machine learning algorithm
The selected algorithm analyzes historical patterns and generates forecasted values for the target version.
The generated data is written directly into the forecast version.
Forecast Inputs
Predictive Forecaster requires historical reference data.
The forecast definition includes:
Forecast Data
This identifies:
- The sheet that receives forecasted data
- The version that receives forecasted data
- Forecast start period
- Forecast end period
Reference Data
This identifies:
- Actuals version
- Historical start period
- Historical end period
The machine learning algorithm uses this reference data when generating predictions.
Supported Data Sources
Predictive Forecaster works with:
- Standard sheets
- Cube sheets
The documentation states that modeled sheets cannot be forecasted directly.
Forecasts can be generated only for supported sheet types and supported account structures.
Historical Data Requirements
Workday documents several requirements for historical data.
One of the most important is the 3:2 ratio rule.
There must be at least three actuals data points for every two forecasted data points.
Example:
- A two-year forecast requires at least three years of actuals data.
Some algorithms require additional minimum historical data points.
The amount of historical data available can affect forecast quality and algorithm selection.
Available Algorithms
Predictive Forecaster includes multiple algorithms.
Examples include:
- AutoFit
- ARIMA
- Croston
- Holt-Winters
- Kalman Filter
- LightGBM
- N-BEATS
- Orbit DLT
- Prophet
Each algorithm is designed for different forecasting scenarios and data patterns. Workday recommends reviewing forecast results and testing different algorithms when necessary.
Rolling Forecasts
Predictive Forecaster also supports rolling forecasts.
When enabled, forecast periods can automatically align with the Completed Values Through date of the selected actuals version.
This allows forecasts to move forward as new actuals data becomes available.
Scheduling Forecasts
Forecasts can be scheduled to run automatically.
Available schedule frequencies include:
- Daily
- Weekly
- Monthly
Organizations often use schedules to keep rolling forecasts aligned with actuals data loads.
Things to Consider
The documentation highlights several considerations:
- Forecast accuracy depends on historical data quality.
- Missing values can affect results.
- Longer forecast horizons are generally less accurate.
- Forecasts overwrite existing data in the selected forecast version.
- Forecasted data cannot be reverted after a successful run.
Predictive Forecaster provides a way to generate forecasted data directly within Workday Adaptive Planning using machine learning algorithms.
The feature supports multiple forecasting methods, rolling forecasts, scheduling, and historical data analysis.
Before implementation, it is important to review data quality, historical data availability, and algorithm requirements to ensure the forecasts generated are appropriate for the planning process.
