Skip to main content
Documentation
Documentation
Adaptive Planning
Last Updated: 2023-06-23
Machine Learning Predictive Forecaster

Machine Learning Predictive Forecaster

With the 2024R2 Release, we introduce Predictive Forecaster, powered by machine learning (ML). Predictive Forecaster leverages your historical data to populate specified forecast versions with ML predicted data.
We enable you to run time series forecasts with regressor data in addition to your historical data within the Workday Adaptive Planning platform. We provide the choice to select the most suitable ML algorithms based on patterns in your data. We also enable you to define the targeted forecast version, time ranges, accounts, and custom dimensions by sheet.
We also provide Confidence Metrics, which enables you to view a dashboard with charts. The charts offer insights into the accuracy and probability range of your forecasts.
Note
: This functionality is only available to Adaptive Planning Limited Preview customers at this time. We plan to deliver this to all customers on September 21, 2024.
What the video:
 6m 52s

Business Benefits

Improve both efficiency and accuracy with Predictive Forecaster. Predictive Forecaster seeds plan versions with ML-generated data, which serves as a reliable starting point for budget managers. Predictive Forecaster uses a science- and math-based approach to forecasting, which can replace or enhance the time-consuming manual process.

Changes

We provide a new
Predictive Forecaster
 permission in the
Permission Sets
 of the
Administration
area. To access Predictive Forecaster, you must add the new permission to your permission sets.
From the new
Predictive Forecaster
 page in
Modeling
, we enable you to create and run forecasts. You can run the forecast or save it and run it later. You can also rerun the same forecast.
In the
Create Forecast
 form, we provide these sections to help you define the forecast:
  • Forecast
    : Specify where to populate the forecasted data. You can select a specific sheet, any unlocked plan version, and the start and end periods.
  • Filters
    : Focus your forecast on specific accounts, levels, and custom dimensions.
  • Reference Data
    : Define the range of historical data that you want the algorithm to study. You can set the start and end periods of an actuals version. We also enable you to add plan versions as additional reference data.
  • Algorithm
    : Select from 8 algorithms. We also enable you to define seasonality and leverage regressor data with lever sheets.
We enable to choose from these algorithms:
  • Prophet
  • N-Beats
  • Holt-Winters
  • Kalman Filter
  • Orbit-DLT
  • LightGBM
  • Croston
  • AutoFit
AutoFit is a machine learning algorithm developed by Workday that automatically picks from the available algorithms based on your specific historical data and trends.
After you save or run forecasts, we enable you to review and sort the list of forecasts. From the list, you can:
  • Access the
    More Actions
     menu.
  • Check the most recent status of each.
  • Link to the sheet that contains the forecasted data.
We display these statuses:
  • Blank: You have saved the forecast definition, but you haven't initiated the 1st run.
  • In Queue
    : You ran the forecast, but it hasn’t started yet.
  • In Progress
    : The forecast is running. We display the percent of completeness.
  • Failed
    : The forecast failed. You can review the errors in the Status section of the Forecast History page and troubleshoot.
  • Success
    : The forecasted data has populated the plan version.
From the toolbar, we enable you to:
  • Delete selected forecasts.
  • Start typing in the
    Search
    field to locate the forecast by name or sheet.
We also enable you to hover over a forecast name and click the 3 dots that display to open the
More Actions
 menu. From the
More Actions
 menu, we provide these options:
  • View Details
    : Review the specifications of the forecast.
  • View History
    : Open the
    Forecast History
    page where we display the status of each run. We also display any error messages of failed runs to help you troubleshoot.
  • Run
    : Begin the process of populating the forecast version with machine learning data.
  • Edit
    : Change the definition of the forecast.
  • Duplicate
    : Start a new forecast with the definition of an existing forecast.
  • Delete
    : Delete the forecast definition — and still retain the populated forecasted data in sheets.
From
Forecast History
 page, we enable you to select a historical run and view the parameters and the status details. In the
Parameters
section, we provide these definition details:
  • Filters
  • Reference Data
  • Algorithm
In the
Status
section we provide these details:
  • How long a successful forecast took to run.
  • A progress bar for pending forecasts that are In Progress.
  • A list of errors for failed forecasts to help you troubleshoot and try again.
For eligible customers, we also provide options that enable you to generate confidence metrics. The
Confidence Metrics
 tab in the
Forecast History
 page enables you to view a dashboard with metrics to give you insights into the accuracy and probability range of the forecast. When you create forecasts, we provide these options:
  • Probability Range
    : Populate forecasted data at the upper and lower limit into 2 additional versions. This option also generates a chart in the
    Confidence Metrics
     that maps the prediction against the upper and lower limits of the prediction.
  • Probability Level
    : Specify how broad or narrow you want the range to be.
  • Accuracy Metric
    : Generate a back-test against existing actuals to get an accuracy metric in the
    Confidence Metrics
     tab.
On the
Confidence Metric
 tab, we enable you to review the resulting charts and filter your perspective.

What Do I Need to Do?

To give users access to the Predictive Forecaster, add the
Predictive Forecaster
 permission to the permission sets.