Predictive AI (Machine Learning)
Trained models that look at your historical data and forecast a number or a probability — the original 'AI in Fusion'.
What it is
Predictive AI — also called classical machine learning — is a model trained on your historical Fusion data to answer one specific question:
"Given everything that's happened before, what's the most likely outcome here?"
It's the oldest form of AI in Fusion and still the most reliable for repeatable, data-rich decisions. Each customer's trained models live in their own securely partitioned tenant — your data never trains anyone else's model.
Where it shows up in Fusion
- EPM Predictive Planning runs Auto Predict to generate a statistical baseline forecast from time-series history. Analysts then refine instead of starting from a blank cell.
- EPM IPM Insights flags anomalies during the close — outlier balances, unusual journal patterns — under time pressure.
- ERP cash-flow forecasting projects collections from open invoices using prior payment behavior.
- HCM time-to-hire and attrition models score candidates and flag retention risk based on historical patterns.
What it is not
It does not generate text, hold a conversation, or take action on its own. It produces a number or a probability. A human (or another agent) decides what to do with it.
When to reach for it
Reach for predictive ML when:
- The decision repeats often (daily, weekly, monthly).
- You have clean historical data (typically 2+ years).
- A number or probability is what you need — not a written explanation.
If you need natural language ("explain why this forecast moved"), that's the next lesson.
Action checklist
Tap each step as you complete it.