Before you Begin

This hands-on tutorial covers the essential tasks for using predictive planning as part of your planning and forecasting cycle.

Background

Use Predictive Planning to predict future performance based on your historical data. You can compare and validate plans and forecasts based on the predictions. For a more accurate and statistically-based forecast, you can copy the prediction values and paste them into a forecast scenario for your plan. Predictive Planning works with EPM Standard and EPM Enterprise applications for Custom and Module application types. For legacy applications, Predictive Planning works with Standard, Enterprise, and Reporting application types. Predictive Planning is not available in FreeForm applications.

What Do You Need?

An EPM Cloud Service instance allows you to deploy and use one of the supported business processes. To deploy another business process, you must request another EPM Enterprise Cloud Service instance or remove the current business process.

  • Have Service Administrator access to an EPM Enterprise Cloud Service instance.
  • Have the Planning Sample Business Process already created.
  • Logged in to the business process with the credentials provided by your organization.

Running Predictive Planning

  1. From the Home page, click the Data card.
  2. Expand the Library folder, then expand the Forecast folder.
    Library and Forecast Folders
  3. Scroll down, then click the Sales Forecast - Products form.
    Sales Forecast Products form
  4. On the form, review sales forecast by product categories for the upcoming planning time intervals.
    5. Form Data
  5. Click the Actions menu on the right side and select Predictive Planning.
    Actions Menu

    When you run Predictive Planning, the program retrieves all the historical data for each member on the form. It then uses sophisticated time series forecasting techniques to predict the future performance for these members. The prediction results are displayed in a panel at the bottom of a form.

  6. Use the down arrow Down Arrow to select the product, Game.
  7. Review the prediction results for the Game product.

    The historical data for this product is shown as a green series on the left side of the chart. The base case prediction is shown in blue on the right. The prediction interval, which is bounded by the Worst and Best cases, is shown as an orange band around the base case prediction.

    Predictive Results for Game
  8. Use the down arrow Down Arrow to scroll through the members on the form, review the historical trends and predictions for each member.

    Compare the forecast for the Software Suite product against the statistical prediction. The Forecast scenario appears on the right side of the chart as a light green series.

    Prediction
  9. Click the dropdown and select Accessories.
  10. On the right side of the panel, view the informational boxes that contain key metrics for each series.

    The Growth Rate metric allows the planner to quickly compare any two series. Judging by these growth rates shown, the forecast is much more aggressive than the statistical prediction, but is still within the prediction interval. The gauge to the right reflects the elevated risk for meeting the sales target for this product.

    11. Accessories Forecast

Understanding the Components of Predictive Planning

Predictive Planning provides a statistically-robust mechanism to help planners create and validate their forecasts using time series forecasting methods on historical data. Most forecasts created by users are based on gut feel or simple growth rates from previous years. However, Predictive Planning allows users to leverage time series forecasting techniques to produce more accurate forecasts.

When you open a form and select a member and then launch Predictive Planning, it produces the following output:

Sales Forecast Products

When you expand the view, you see the following:

Form with Callouts
  1. Member selection dropdown: Use to select any member on the form and explore the predictive planning output.
  2. Chart area: Shows the data for the chosen member. On left side of the chart, it has the historical actual data. And, on the right side (partitioned by the vertical line), it shows the forecast and prediction for the future time horizon. It also contains the best case (optimistic scenario) and worst case (pessimistic scenario).
  3. Historical Data Details: Provides information on the historical data used for running the forecast algorithms. It includes the number of historical observations, missing values, outliers, presence of seasonality, etc.
  4. Prediction Details: This section provides details on the prediction output for the best-performing algorithm. Predictive Planning runs a set of time series forecasting algorithms on the historical data and picks the output from an algorithm that gives the best accuracy for the given member. It shows the name of the algorithm that has the highest accuracy compared to other algorithms and it provides RMSE and Accuracy metrics.
  5. Info boxes: The info boxes on the right side provide a statistical summary of each series and are always visible on the right side of the chart, normally one box per series. The order of the boxes matches the order of the series in the legend.
    1. Growth Rate statistic is provided in each box as the key metric for comparing one series against another.
    2. Risk Gauge is added next to the growth rate to indicate the probability of the scenario occurring above or below the prediction.

How Predictive Planning Works

Predictive Planning is accessible from any form via the Actions menu and it retrieves all the historical data for each member on the form. It then uses sophisticated time series forecasting techniques to predict the future performance for these members.

Forecasting Algorithms

Two primary techniques of classic time-series forecasting are used in Predictive Planning:

  • Classic Non-seasonal Forecasting Methods — Estimate a trend by removing extreme data and reducing data randomness
  • Classic Seasonal Forecasting Methods — Combine forecasting data with an adjustment for seasonal behavior

Method Seasonal Best Use
Simple Moving Average No Volatile data with no trend or seasonality
Double Moving Average No Data with trend but no seasonality
Single Exponential Smoothing No Volatile data with no trend or seasonality
Double Exponential Smoothing No Data with a trend but no seasonality
Damped Trend Smoothing non-seasonal method No Data with a trend but no seasonality
Seasonal Additive Yes Data without trend but with seasonality that does not increase over time
Seasonal Multiplicative Yes Data without trend but with seasonality that increases or decreases over time.
Holt-Winters’ Additive Yes Data with trend and seasonality that does not increase over time
Holt-Winters’ Multiplicative Yes Data with trend and with seasonality that increase over time
Damped Trend Additive Seasonal Method Yes Data with a trend and seasonality
Damped Trend Multiplicative Seasonal Method Yes Data with a trend and with seasonality
ARIMA No Data with minimum of 40 historical data points, limited number of outliers and no seasonality
SARIMA Yes Data with minimum of 40 historical data points, limited number of outliers and seasonality

All of the non-seasonal forecasting methods are run against the data. If the data is detected as being seasonal, the seasonal forecasting methods are run against the data.

Selecting the best-performing forecasting model

The forecasting method with the lowest error measure (RMSE) is used to forecast the data. RMSE (root mean squared error) is an absolute error measure that squares the deviations to keep the positive and negative deviations from cancelling out one another. This measure also tends to exaggerate large errors, which can help eliminate methods with large errors. For instance, the forecasts from multiple algorithms are compared against each other based on the RMSE. The forecasting model with the lowest error, i.e. RMSE is chosen the best by default.

List of Methods

Example – Forecasting non-seasonal data

Here, we’re looking at the prediction results for Sales by product categories for the “Sales East” entity.

Products Form

Select the “Software Suite” product. View the prediction output at the bottom panel. The historical data for the Software Suite product category is shown as a green series on the left side of the chart. The base case prediction is shown in blue on the right. The prediction interval, which is bounded by the Worst and Best cases, is shown as an orange band around the base case prediction. The historical data seems to be on an increasing trend and there is no obvious seasonality. Click on the Info icon at the top right corner on the panel, to see more information about the prediction.

Details

The prediction output for this product category is from the “Damped Trend Non-Seasonal” method as it has the lowest error measure (RMSE) of 154. The prediction has an accuracy of 90% which is the likelihood of happening.

Prediction Details

Example – Forecasting seasonal data

Here, we’re looking at the prediction results for the Monitor product in the “Sales East” entity. The historical data for the Monitor product category is shown as a green series on the left side of the chart. The base case prediction is shown in blue on the right. The prediction interval, which is bounded by the Worst and Best Cases, is shown as an orange band around the base case prediction.

Best Prediction Method

The historical sales for the Monitor product category have been seasonal as it hits the peak around August and sees lowest sales in January every year. The “Seasonal ARIMA” method produces the most accurate results for this product category. Interestingly, the chart also captures the seasonality in the data as “seasonal bands”.

Seasonal Data

Example – Forecasting seasonal data without trend

Below, view the prediction results for the Accessories product in the “International Sales” entity. The historical data for Accessories product category is shown as a green series on the left side of the chart. The base case prediction shown in blue on the right. The prediction interval, which is bounded by the Worst and Best cases, is shown as an orange band around the base case prediction.

Accessories

The historical actual sales show seasonality but no visible trend. The “Seasonal Multiplicative” method gives the most accurate results for the given scenario.

Example – Forecasting non-seasonal sales with large historical data

Below, view the prediction results for the Keyboard product in “Sales East” entity. The historical data for Keyboard product category is shown as a green series on the left side of the chart.


Keyboard

The historical actual sales show no seasonality but there is a clear increasing trend. Since it has a good amount of historical data (more than 40 data points), the “ARIMA” method gives the most accurate results for the given scenario.

Example – Forecasting seasonal sales with large historical data

Below, view the prediction results for the Monitor product in the “Sales East” entity. The historical data for the Monitor product category is shown as a green series on the left side of the chart.

Monitor

The historical actual sales show seasonality and there is also a clear increasing trend. Since it has a good amount of historical data (more than 40 data points), the “Seasonal ARIMA” method gives the most accurate results for the given scenario.

Changing Settings in Predictive Planning

Once Predictive Planning is run, view the prediction output at the bottom of the page in the panel. You can see the default settings used for the prediction, which you can configure or customize as needed.

  1. Run Predictive Planning on the Sales Forecast – Products form. Click the Settings icon at the right corner of the panel.
    Settings
  2. In Settings, click the Linear Trend Line (Past) checkbox.
    Linear Trend Line

    The trend line for historical sales for the Television product has been added to the chart.

    Trend Line
  3. Select the Game product for Sales East entity. Notice that it has a declining trend in sales.
    Game Trend Line
  4. In Settings, select the Date Ranges tab.
    Date Ranges
  5. Using the drop-down list, change the Future End Period from Jun to Sep, then click Apply.
    Future End Period

    Notice the prediction/future time horizon is extended by 3 months, up to September.

    Future End Period

Adjusting the Forecast Based on Prediction

Once predictions have been calculated using Predictive Planning, compare the current forecast scenario with the predictions and make adjustments where required. This can be done by manually adjusting the forecast series by comparing it with the prediction.

  1. Open the sales forecast form and trigger the Predictive Planning calculations using the Actions menu as described in the previous topics. View the predictions for each member on the form.
    2. Sales Products
  2. Scroll through the members on the form and view the historical trends and predictions for each member. Then, select Network Card on the form to view the forecast for the Network Card product against the statistical prediction.
    Network Card

    The prediction for Network Card product seems to be lower than the forecast scenario. We can adjust the forecast downward. We’ll first zoom into the prediction range and adjust the forecast manually by dragging the series on the chart.

  3. Click Zoom in to see the right side portion of the chart in expanded view.
    4. Zoomed In
  4. View the expanded graph:
    5. Expanded View
  5. Adjust the Forecast series on the form by manually dragging the Forecast (Working) line on the chart below.
    Drag forecast line

    Note that as you drag the forecast line on the chart, the numbers are updated on the form.

Pasting the Prediction into the Forecast

Once the predictions are calculated using Predictive Planning, you can compare the current forecast scenario with predictions and make adjustments as needed. In the previous topic, you saw how this can be done by manually adjusting the forecast series by comparing it with the prediction. Alternatively, you can copy the prediction series and paste it to the forecast series as required.

  1. Open the Sales Forecast form and run the Predictive Planning calculations using the Actions menu.
    Sales Prod Form
  2. Scroll through the members on the form and view the historical trends and predictions for each member. Select Game on the form and view the forecast for the Game product against the statistical prediction.
    Game Product
  3. Click the Paste icon on the top right corner of the Predictive Planning chart panel.
    Game Paste
  4. In Paste Prediction, select the range and series where you want to paste the predictions, then click Paste.
    Paste

  5. View the prediction results pasted into the Forecast scenario for the Game product, then click Save.
    Save Changes

Best Practices

The following are some of the best practices on leveraging the benefits of Predictive Planning.

  1. Historical Data: Historical data for at least twice the number of prediction periods is needed. For example, to predict 12 months in the future you should have at least 24 months of past data. The more historical data, the more accurate the prediction. At the time of prediction, if there is not enough historical data available, a warning is displayed.
  2. Seasonality: At least two complete cycles of data must be available to detect seasonality.
  3. Time Granularity of Predictions: On the form include the lowest level of Period members possible so the greatest amount of historical data can be used. This is because the lowest Period dimension member level on a form determines the time granularity of the prediction. For example, if the lowest member level on the form is Quarters, the historical data will be retrieved at the Quarters level and the prediction will also take place at the Quarters level.
  4. Daily Forecasts: If there are gaps in the historical data for weekend days there will be issues forecasting daily. Instead, aggregate historical data and forecast to a weekly or monthly level.
  5. Events: Predictive Planning does not support events; i.e. identifiable occurrences that have affected historical data and could affect predicted data. Examples are recurring events like marketing promotions, one-time events like natural disasters or recurring events with moving dates like Ramadan. Adjustments for these events must be handled manually after the prediction is made.
  6. Prediction Scenario: On your form, include a Prediction scenario for pasting and saving the results of the prediction. This enables you to later compare the historical predictions against the historical forecasts to determine the accuracy of each.
  7. Expand all columns and rows before adjusting values: When you are pasting predicted values into a scenario, or you are changing form values by dragging the chart, all columns and rows should be expanded. Note that you can't paste values to ad hoc grids and composite forms.

Additional Resources

RESOURCE DESCRIPTION
Video: Overview: Statistical Forecasting Methods in Predictive Planning for Planning and Budgeting Cloud Video on Predictive Planning and forecasting methods
Documentation: Chapter 10 Improving Forecasting Accuracy with Predictive Planning from Working with Planning for Oracle Planning and Budgeting Cloud Documentation for using on web

Includes links to overview videos. Last topic includes link to video on using with Smart View
Technical Paper: Predictive Planning Forecasting and Statistical Descriptions Documentation on forecasting methods
Documentation: Working with Predictive Planning in Smart View Documentation for use with Smart View.

Last section has details on the forecasting methods (same details as in Predictive Planning Forecasting and Statistical Descriptions)

Learn More