How Oracle Cloud EPM Uses AI to Transform Forecasting - Auto Prediction

 Forecasting has always been at the heart of enterprise planning. Yet, traditional approaches depend heavily on manual analysis, spreadsheet logic, and intuition—often resulting in delays, inconsistencies, and limited visibility into future outcomes.

With the rise of EPM AI, Oracle Cloud EPM has transitioned into a modern, intelligence‑driven forecasting engine. Today, planners can rely on Predictive AI, Generative AI, and the emerging Agentic AI to generate, interpret, and enhance forecasts with remarkable accuracy.

This blog is part of a series exploring one of the most powerful capabilities in EPM AI: Forecast Insights—enabled by Predictive Planning, Auto Predict, IPM Insights, and the latest Advanced Predictions

Oracle AI Journey (Oracle - AI Features )

Started with Predictive Planning (2016).

It expanded with Auto Predict (2020).

It grew deeper with IPM Insights (2021).

It took a major leap in 2025 with Advanced Predictions, introducing multivariate machine learning into EPM Cloud.

This evolution is validated through Oracle’s EPM AI feature listings and historical timelines 

These advancements reflect Oracle’s strategy to deliver AI embedded natively throughout EPM, not as external or disconnected tools. 

Oracle EPM Agent AI Studio in future release

This blog is part of a series exploring one of the most powerful capabilities in EPM AI: Forecast Insights—enabled by Predictive Planning, Auto Predict, IPM Insights, and the latest Advanced Predictions.

Auto Predict 

Oracle EPM automatically analyzes historical data, selects the best forecasting model, and generates reliable future projections—helping finance teams create data‑driven forecasts quickly and efficiently.

Auto Predict is typically used for:

• Revenue forecasting
• Expense trend projection
• Demand planning
• Cash flow forecasting
• Baseline generation for forecasts

It works best when:

• You have at least 12–24 periods of historical data
• Data follows a time-based pattern
• Minimal manual overrides exist

How it works?

At a high level - It's a 5 Step process 

1. Data Profiling 

• Analyzes historical patterns.
• Identifies seasonality, trend, volatility
• Detects missing values and anomalies

2. Model Evaluation

• Naive / Baseline models
• Simple Moving Average (SMA)
• Exponential Smoothing
• ARIMA-based models

3.Model Scoring

• AIC (Akaike Information Criterion)
• AICc (Corrected AIC)
• BIC (Bayesian Information Criterion)

4. Best Model Selection

•     Model with lowest information criterion is stored.

5. Prediction Generation 

• Forecast values are written back to EPM Cube.

As you see in the above picture, Model with least AIC number is selected. ARIMA (1,0,0) means

• AR (1) - One prior period explains the future value
• No differencing (0): Data is already stationary
• No moving average (0): Noise is minimal

We don't have to be a data scientist but if we know our data well then we can validate the report generated by the Oracle. Auto Predict cannot give you the final forecast version but it can be the starting point based on your historical data.

Few more Model keywords

1. Trend - Underlying direction of the data over time. For instance, Upward sales growth → positive trend.

2. Gap - Data is missing for certain periods.

3. Seasonality - Regular, repeating pattern in data that occurs at predictable intervals. For instance, Higher retail sales every December or increased travel costs during summer.

4. Outlier - Unusual values that don’t reflect normal behavior. Auto Predict may adjust or smooth outliers internally so they don’t overly influence the forecast

5. Events - One‑time or unusual business occurrences that cause a temporary spike or drop in data.

That’s a lot of concepts to take in, so let me pause here and move on to the steps involved in configuring this powerful feature.

Steps to create the Auto Predict

1. Navigate to IPM > Configure > Create

2.  Give a name for your Auto Predict. Select the Auto Predict option under General Predictions

3. Click on Next

4. Select a Calendar if you have but for our demo I have not created any Calendar as it's not mandatory for Auto Predict.

5. Pick the Cube for historical period and Forecast period.

6. Click Next. Select the Historical POV and Future Base Prediction POV.

7. Click on Show Advanced Predictions to view the additional options available.

    Note : For now, no changes are required. If you choose to experiment with these settings, make sure you review the key terms and explanations covered earlier in this blog to understand their impact.

8. Click on Save

9. Click on Actions > Run.

10. After run is complete. Actions > Download Report.

 11.  Report has 3 tabs 

• Summary
• Gives the summary of the Insight configuration
• Report
• Gives the details of the Prediction Statistics for each account members and their accuracy details.
• SV_Series_Data_OEP_FS
• Shows the POV of the data, which can be used to create the SV templates.