Recently, I was able to test out the IPM tools in EPBCS. We utilized Auto Predict and all three of the IPM insights. While the tool proved to be surprisingly useful, we encountered a few challenges throughout the testing.

We used:

• Auto Predict: Improves the accuracy of forecasting.
• Forecast Variance and Bias Insights: Reveals hidden biases in forecast or plan submissions by analyzing historical data.
• Prediction Insights: Identifies significant deviations in forecasts or plans compared to predicted values.
• Anomaly Insights: Detects unusual patterns in data that deviate from expected results.

For this blog, I will focus solely on Auto Predict. Auto Predict automatically picks the best predictive algorithms based on the historical data available. The selection process is driven by performance metrics, like Root Mean Squared Error (RMSE), to make sure the chosen model fits the data’s unique characteristics. Users have the option to apply seasonal methods or non-seasonal methods and also have the option to choose ARIMA or Extend ARIMA to predict the data. Extend ARIMA option will take significantly longer to predict data; therefore, ARIMA should be the default selection.

The user options are limited but Auto Predict function is still very useful. It can be run on the intersection you define, unlike Predictive Planning, which is limited to running predictions on a form. However, despite its usefulness, we encountered some challenges:

• Data inconsistency
• Data sparsity in the Period dimension
• Meaningless predictions for certain accounts at Lv0, such as Headcount

To address these challenges, we created a new alternative hierarchy and added “_IPM” members as “Store”. We then copied the parent-level data to the “_IPM” members for most of the dimensions, except for those that we could define at Lv0. Once we did this, we were able to obtain much more consistent and meaningful prediction data for referential purposes.

Sparsity still exists here and there but we are getting much better data. Now, what about data promotion to Lv0 to pre-populate the Forecast/Plan?

We asked the users to help us understand the correlation between Actual/Final data and their Forecast/Plan data. We understand that the users are only using certain accounts for planning purposes, so we created a DATACOPY rule to copy the data from the “_IPM” members to the specific plan accounts/entities and a few other custom dimensions. This is a rather complicated and a not so creative approach but it does the job for the time being.

Overall, I am impressed with the Auto Predict capability, but there is room for improvement. For example, all predictions are time-series based, meaning they utilize historical data to predict future data. Time-series prediction assumes past behavior will continue in the future, which is not always the case with financial data or KPIs (such as Headcount, FTE, etc.). Oracle plans to introduce multivariate prediction in the near future, which will be very useful for predicting rates, assumptions, drivers, and more. Let’s see how beneficial it will be when it’s released!