The Future of FP&A Intelligence
From reporting performance to understanding drivers, forecasting continuously, and shaping decisions.
For a finance team moving from Excel toward EPM, the objective is not simply to put the same workbook on a new platform. It is to create a planning process in which assumptions, ownership, and decisions remain understandable as the organization grows. AI can contribute to that process, but it does not remove the need for a sound financial model.
What the current EPM direction tells us
Oracle’s current EPM guidance describes predictive planning, anomaly detection, generative narratives, assistants, and agentic workflows. It recommends choosing an initial AI use case with bounded complexity and clear value. This is vendor guidance about its platform; feature availability and suitability must be checked for the customer’s actual environment.
Source: Oracle — Selecting the First EPM AI Use Case (Documentation accessed September 12, 2026).
Connect the plans before automating the commentary
For an illustrative global business, revenue, expense, and workforce planning are connected. Hiring assumptions affect compensation expense; revenue assumptions may influence staffing needs; currency choices affect the view presented to group finance. The implementation should make these relationships explicit rather than hide them inside separate spreadsheets.
Start with a shared definition of entities, accounts, periods, scenarios, and ownership. Document which assumptions local teams may change and which are controlled centrally. Preserve the business reasoning behind important workbook calculations so that migration does not silently change the planning logic.
Measure three different improvements
A timely close, an accurate forecast, and a useful variance explanation are different outcomes. They should not be compressed into one vague “finance efficiency” measure. Track the time required to complete the relevant process, the agreed forecast-error measure, and the effort needed to investigate and approve exceptions.
Keep the numerical calculation of a variance separate from its explanation. An AI-generated narrative may propose a driver to investigate, but a plausible sentence is not evidence of causation. Reviewers need access to the underlying periods, assumptions, and transactions or operational measures that support the explanation.
Use the preparation window well
If implementation is expected to start in three or four months, use the intervening period to establish source-system access, identify finance process owners, inventory important models, and select a representative first planning cycle. Distinguish the implementation start from the first production release.
A useful initial scope may be a planning process that can be reconciled and signed off end to end. Expansion should follow evidence that the model behaves as intended and the business can operate it. Agree how unresolved data issues, model changes, and new requirements will affect scope before the rollout grows.