Your team is being asked to do something with AI. The pressure is real, it’s coming from the board, and it rarely arrives with a clear definition of what “something” means.
Meanwhile, Oracle has spent the last several years building AI directly into Cloud EPM — not as a separate product, but as features sitting inside the planning, forecasting, and reporting work your team already does. Most finance teams don’t know these tools exist. The ones who do aren’t sure whether to trust them. And the gap between the two is where good decisions get made or missed.
Here’s a plain account of what’s there, what it’s good for, and where to start. No hype. We’re learning this alongside our clients, and we’d rather tell you what we’ve actually found than sell you a transformation.
What’s Actually in the Platform
The capabilities live under Intelligent Performance Management — IPM. Four are worth knowing.
Auto Predict produces a statistical forecast from the history of a single measure, and can run across thousands of members as a scheduled job to seed a cycle before anyone touches it. Oracle shipped it in 2020. It’s well-proven, fast, and useful as a baseline when the data is clean.
IPM Insights reads your data for anomalies, forecast bias, and variances a reviewer would miss, and flags what’s behaving differently than expected. It’s been available since 2021. Think of it as a second set of eyes on a large model.
Advanced Predictions, new in August 2025, moves from one variable to many — using machine learning to weigh drivers like price, volume, and exchange rates together. It requires the EPM Enterprise tier and has to be enabled, so confirm your edition before you count on it.
Generative AI and agents are the newest layer, and Oracle is still rolling them out. Drafted narrative commentary and conversational agents that answer questions in plain language are promising — but availability depends on region and subscription, and Oracle itself frames them as decision support to be reviewed, not trusted blindly.
None of this needs a data science team. Oracle built it for finance users, in their words, “without needing to learn data science.” If you run Cloud EPM, you likely already own most of it.
What Works, and What to Watch
The forecasting tools work — within limits, and the limits matter.
They are strongest where the data is clean, consistent, and deep enough to learn from. Oracle is candid about this: Auto Predict wants at least twice as much history as the periods you’re predicting, and it drops to a straight line when the history has no real pattern. A model trained on the past can’t see a future that doesn’t resemble it. That isn’t a flaw. It’s the math.
So the useful posture is neither excitement nor suspicion. It’s judgment. These tools produce a defensible baseline fast, which frees your team to spend its time where the work actually is — the drivers, the assumptions, the scenarios that need a person. They don’t replace that work, and anyone telling you they do is selling something.
Anomaly detection earns its place first. Surfacing “this looks wrong, look here” assists a person without asking them to surrender a decision — which is why it pays off under the pressure of close. The generative features, where you have them, draft a starting point, not a final answer. Oracle’s own guidance is to review the output before it’s used. That’s the right instinct for all of it.
The Real Constraint Is Underneath
Most teams treat AI in EPM as a technology decision. It isn’t. The constraint is the data and the architecture beneath the features.
Point Auto Predict at messy hierarchies, inconsistent metadata, and actuals full of one-time noise, and it will hand you confident forecasts that aren’t worth much — because it faithfully learns whatever you give it. The old principle holds: a model is only as good as what it learns from.
For a CFO, that’s good news. The path to value from AI runs straight through work worth doing anyway — a clean planning architecture, governed data, and a forecasting process your people understand and trust. AI raises the return on that discipline. It doesn’t excuse skipping it.
Where to Start
We’d suggest a sequence, not a leap.
Start with an honest read of where you are. Where does your team lose time to manual, repeating work? Where is forecast accuracy a problem worth solving? Where is the data clean enough to trust, and where isn’t it? Those answers matter more than which features you switch on — and they’re worth checking against your Oracle edition, since the newer tools need the Enterprise tier.
Then move where the risk is low. Anomaly and variance flagging help without asking anyone to hand over judgment. Treat baseline forecasting as a draft — let Auto Predict seed it, and put your team’s hours into the assumptions on top. And give the newest layer time; let the generative and conversational tools prove themselves on lower-stakes work while you build the foundation that makes everything above more reliable.
The Question Worth Asking
The pressure to “do something with AI” pushes organizations to adopt tools before doing the quiet work that makes those tools pay off. We’d start somewhere else: where is your team spending real effort on work that takes little judgment? That’s where AI in EPM earns its place — not by transforming finance overnight, but by clearing friction from work your team already understands, so their attention goes to the decisions that move the business.
Oracle has built capable tools and keeps adding to them. On a sound foundation, with judgment, they’re worth your attention. As a shortcut around the fundamentals, they disappoint. The difference was never the technology. It’s the thinking around it — and that part has always been the work.