Selecting the First EPM AI Use Case
The current EPM AI feature set — spanning predictive planning, anomaly detection, generative AI for narratives, AI Assistants, and agentic workflows — is documented in Features with AI. This documentation is updated regularly and is the authoritative reference for available features.
Use Case Selection Framework
The following framework positions Oracle Fusion Cloud Enterprise Performance Management AI use cases by business value and implementation complexity. Select the first pilot from the high value / controlled complexity category.
| Category | Definition | EPM process Examples |
|---|---|---|
| High value · Controlled complexity (start here) | Processes with significant business value, a bounded decision scope, reliable inputs and a clear success metric. These are the appropriate starting point. | Anomaly detection in EPM processes (IPM Insights), Variance commentary generation (Narrative Reporting GenAI); Account reconciliation guidance (Account Reconciliation Assistant) |
| High value · High complexity (do later) | High-value processes that require complex automation or heavy cross-system orchestration. Appropriate at Stage 3–4, after the CoE has demonstrated production deployment capability. | Predictive multi-driver planning models with external data feeds; Cross-process AI orchestration; Autonomous close workflows |
| Lower value · Lower complexity (quick wins) | Useful for building AI literacy and early momentum. Do not treat these as the destination — move to higher-value processes as soon as the CoE is ready. | Knowledge lookup and guided reporting support; Process navigation guidance |
| Lower value · Higher complexity (avoid) | High implementation cost with limited measurable return. Deprioritize until higher-value use cases are proven. | Complex data integration with no direct finance output; Broad AI strategies without a process anchor |
Five Qualification Questions
Each of the following questions must have a specific answer before a use case is approved for implementation. Vague answers at this stage produce scope problems, governance gaps and failed deployments.
- Who owns the finance process? A named finance professional — not IT, not the CoE — who will act on the AI output and is accountable for its accuracy.
- What data, documents, or tools does the agent require? A defined, governed and currently available set — not data that must be gathered or cleaned before the pilot begins.
- What actions can the agent take without human approval? A short, explicit list. All actions not on this list require human review by default.
- What must remain human-approved? All actions with material financial impact, audit implications or regulatory consequence.
- Which metric proves pilot value? A specific measurable quantity with a known baseline — not a qualitative assessment.