Understanding Predictive Forecaster in Workday Adaptive Planning

Forecasting is one of the core parts of any planning process. Finance teams use historical data, current business trends, spreadsheets, driver assumptions, and management input to estimate future results. Workday Adaptive Planning includes Predictive Forecaster to support this process by generating forecasted values from historical reference data.

Predictive Forecaster should not be treated as a replacement for the planning process. It is better understood as a forecast seeding capability. It can create a first version of forecast data that budget owners and FP&A teams can review, adjust, and challenge before it becomes part of the final plan.

What Predictive Forecaster Does

Predictive Forecaster is an Intelligent Planning capability in Workday Adaptive Planning. It uses historical data to generate forecasted values and writes those values into a selected plan version.

From a planning architecture perspective, this is important. The forecast output is not just a report or an external file. The generated values are written back into Adaptive Planning, which means they can be used directly in sheets, reports, dashboards, and planning workflows.

The practical use case is simple. Instead of starting a forecast version with blank values or manual copy-forward assumptions, Predictive Forecaster can seed the version with system-generated forecast values. Finance users can then review those values and make adjustments based on business knowledge.

How the Forecast Process Works

The process starts by defining what needs to be forecasted and what historical data should be used as the reference.

A user selects the sheet where forecasted values will be written, the target forecast version, the forecast start and end periods, the historical actuals version, the historical start and end periods, and the forecasting algorithm.

Once the forecast is executed, Predictive Forecaster analyzes the historical reference data and generates forecasted values for the selected forecast period. Those values are then written into the selected forecast version.

This setup makes the forecast process controlled. The user defines the source history, the target version, the forecast period, and the method. That control matters because a forecast should not be generated without knowing exactly where the data came from and where the output will be stored.

Forecast Data and Reference Data

A Predictive Forecaster setup has two important data areas: forecast data and reference data.

Forecast data defines the output side of the process. It identifies the sheet that will receive the forecasted values, the version where the values will be stored, and the forecast start and end periods.

Reference data defines the historical input side of the process. It identifies the actuals version and the historical period range that the selected algorithm will use to generate the forecast.

This distinction is important during implementation. The forecast version and actuals version serve different purposes. Actuals provide the historical pattern. The forecast version receives the generated output.

If these are not defined clearly, users may misunderstand the process or accidentally write forecast values into the wrong planning version.

Supported Sheet Types

Predictive Forecaster supports standard sheets and cube sheets. These are common sheet types used for planning and reporting in Workday Adaptive Planning.

Modeled sheets cannot be forecasted directly using Predictive Forecaster. That is an important design consideration because many Workforce, Revenue, Capex, and operational planning models may use modeled sheets. If the planning process relies heavily on modeled sheets, the team needs to review where Predictive Forecaster fits and whether the forecast should be applied to related cube or standard sheet structures instead.

This should be confirmed during design before users assume every planning area can be forecasted directly.

Historical Data Requirements

Predictive Forecaster depends on historical reference data. The quality and volume of that history directly affect the usefulness of the forecast.

One key requirement is the historical-to-forecast data ratio. There must be enough historical actuals to support the forecast horizon. A common rule referenced in the documentation is the 3:2 ratio, meaning there should be at least three actuals data points for every two forecasted data points. For example, a two-year forecast would require at least three years of actuals data.

Some algorithms may require additional historical data points. This means algorithm selection is not only a user preference. It depends on the available data, the forecast horizon, and the pattern in the historical results.

In practical terms, a longer history usually gives the forecast process more context. But history should also be relevant. If the business has changed significantly, old history may not represent the current operating model.

Forecasting Algorithms

Predictive Forecaster includes multiple forecasting algorithms, including AutoFit, ARIMA, Croston, Holt-Winters, Kalman Filter, LightGBM, N-BEATS, Orbit DLT, and Prophet.

Each method is designed for different data patterns and forecasting scenarios. Some methods may handle trend and seasonality better. Some may be more appropriate for intermittent or irregular data. Others may perform better when there is enough consistent history.

For most FP&A users, the key point is not to memorize every algorithm. The practical approach is to test the forecast output, compare results, and select the method that produces the most reasonable forecast for the planning use case.

AutoFit can be useful as a starting point, but the forecast output still needs review. A technically generated forecast is not automatically a good business forecast.

Rolling Forecast Support

Predictive Forecaster can also support rolling forecast processes. When rolling forecasts are enabled, forecast periods can align with the Completed Values Through date of the selected actuals version.

This is useful for organizations that refresh actuals regularly and want the forecast period to move forward as new actuals become available. For example, after a new month of actuals is loaded and completed, the forecast can be generated for the next open periods.

From a process perspective, this fits well with monthly forecast cycles. Actuals are loaded, the actuals period is closed or completed, and the forecast is regenerated for the remaining planning periods.

Scheduling Forecast Runs

Forecasts can also be scheduled. Supported schedule frequencies include daily, weekly, and monthly.

Scheduling can be useful when the organization has a repeatable process and the source data is updated on a known cadence. For example, a monthly schedule may be appropriate after actuals are loaded and validated.

However, scheduling should be used carefully. Since forecasted values are written into a target version, the team should confirm that the version is ready to receive updated forecast data before the job runs. A scheduled forecast should not overwrite user-reviewed planning data unexpectedly.

Important Implementation Considerations

Before enabling Predictive Forecaster, the team should review the source data, version design, forecast horizon, sheet structure, and business process.

Data quality is one of the most important factors. Missing values, inconsistent history, one-time events, business restructures, and changes in account or level structure can all affect the forecast output.

The forecast horizon also matters. Near-term forecasts are usually easier to support than long-term forecasts. As the forecast extends further into the future, uncertainty increases.

The team also needs to understand that forecast runs can overwrite existing data in the selected forecast version. This is a major implementation point. The target version should be protected, named clearly, and used intentionally. Users should not run forecasts into a working version without understanding the impact.

Because forecasted data may not be easily reverted after a successful run, version backup and process controls should be part of the design.

How FP&A Teams Should Use Predictive Forecaster

The best way to use Predictive Forecaster is as a starting point for planning review.

The generated forecast can help seed a version, reduce manual starting-point work, and give users a data-driven baseline. FP&A teams can then compare the output against business expectations, known changes, pipeline, headcount plans, seasonality, and management guidance.

This is where finance judgment remains important. Predictive Forecaster can identify patterns in historical data, but it does not know every future business decision. It may not know about a new product launch, a pricing change, a restructuring, a major customer loss, or a one-time investment unless those items are reflected in the planning model.

The forecast should be reviewed, not accepted blindly.

Final Observation

Predictive Forecaster in Workday Adaptive Planning provides a structured way to generate forecasted values directly inside the planning system. It supports standard and cube sheets, uses historical reference data, offers multiple forecasting algorithms, supports rolling forecasts, and can be scheduled.

From an implementation perspective, the most important points are data quality, version control, historical data availability, algorithm selection, and user review. The feature can add value when it is aligned with a clear planning process and when users understand how the forecast output should be used.

Predictive Forecaster is not a complete planning process by itself. It is a tool that can help seed forecast values and support better planning conversations. The final forecast still needs business review, ownership, and governance.