Before the Agents: How Data Integration, EDM, and Metadata Alignment Make (or Break) AI in EPM
An EPM Intelligence podcast recap featuring Mike Casey, Oracle’s Product Manager for Data Integration (EPM).
Key Takeaways
- AI output in EPM is only as trustworthy as the metadata underneath it.
- Oracle’s AI strategy spans native agents, custom agents, and open interoperability.
- Chart of accounts alignment and EDM are the real starting point for AI in EPM.
- Done well, integration should be invisible to the CFO, running quietly in the background while AI frees up time for more complex tasks.
Why AI in EPM Starts with Integration, Not Agents
In our EPM Intelligence podcast episode with Mike Casey, who has led data integration for Oracle Enterprise Performance Management (EPM) for nearly three decades, shares what makes AI in finance reliable. While the market is focused on demoing finance agents that forecast, reconcile, and write narratives, Mike makes the case that none of that output is trustworthy without clean metadata, aligned charts of accounts, and the integration work that happens well before an agent ever runs.
This article explores how Oracle is structuring AI across EPM, why chart of accounts alignment and metadata are the real foundation for AI readiness, and what a practical roadmap looks like for finance teams getting started.
Oracle’s Three-Part AI Strategy
Mike frames Oracle’s approach to AI in EPM around three distinct layers, all sitting on top of a shared governance framework.
- Native Agents and Assistants: Built directly into the application, these include tools like an account reconciliation assistant and a consolidation assistant that walks users through closing the books conversationally.
- Custom Agents via AI Agent Studio: For organizations that need more than what’s available out of the box, AI Agent Studio lets teams build their own assistants tailored to specific processes.
- Interoperability: An open API path lets external models plug into EPM, so organizations aren’t locked into a single AI provider.
Predictive machine learning and analytics, including capabilities inside IPM Insights, run alongside these layers. Most environments are still at the assistant stage, where agents perform actions similar to a knowledgeable teammate working alongside a human.
Choosing an LLM: Token Costs and Output Quality
One practical question finance and IT leaders are asking is: What does it actually cost to run AI at scale?
Mike notes that model selection carries a usage component: as organizations consume more AI, that consumption translates into a cost model, typically structured around credits, that finance teams will need to plan for as usage evolves.
Cost isn’t the only variable. Output quality can also differ meaningfully between LLMs, and the “right” model can depend on the type of question being asked. For organizations building an AI strategy in EPM, this means model selection deserves the same scrutiny as any other platform decision.
The Real Foundation for AI: Chart of Accounts Alignment and EDM
If there’s one technical detail that determines whether AI in EPM is trustworthy, it’s chart of accounts alignment between the ERP and EPM platform. Without consistent metadata across systems, there’s no reliable way to reconcile what an account or KPI means in EPM against what it represents at the source.
What Happens When Metadata Isn’t Aligned Upfront
When an AI agent generates an answer in EPM, a user often needs to drill back to the ERP to validate it. If the metadata isn’t aligned, that drill-through breaks, and the result is more than an inconvenience. It erodes trust in the agent’s output entirely. Misaligned metadata is, in Mike’s words, one of the first problems organizations run into when they jump into AI across ERP and EPM domains without addressing alignment first.
EDM as the Single Source of Truth
Enterprise Data Management (EDM) is central to making this alignment achievable at scale. EDM defines what data structures look like on both the ERP and EPM sides and manages how they map to one another, creating a single source of truth that AI models and agents can rely on. The transformation work has to happen somewhere. Without EDM front-loading that mapping upstream, organizations end up handling translation logic downstream inside EPM instead.
Chart of accounts alignment is a prerequisite, and organizations planning an AI rollout should treat metadata alignment as the first project milestone, not a parallel workstream.
A Customer Example: Cleaning Data Before It Loads
Mike described a customer running roughly 200 ERP source systems into EPM that built an AI-driven pre-load validation process. Before submission, the system pulled the mapping rules, ran incoming data against them, identified exactly where records would fail, and generated corrected mapping rules for review, all before the data ever hit EPM. The result was a cleaner load, with few errors to troubleshoot after the fact.
Native Oracle AI Agent or Build Your Own Agent?
As AI capabilities expand, organizations have to decide when to use functionality Oracle already provides and when to build something custom through AI Agent Studio. Mike’s advice is to start with the problem, not the tool. Define what you are trying to solve, then determine whether an existing Oracle agent, assistant, or predictive analytics capability already addresses it before investing time in a custom build that may later become native functionality.
AI capabilities are already available across areas including:
- Account Reconciliation
- Financial Consolidation
- Profitability
- Reporting
- Predictive analytics and IPM
* Mike also highlighted the Planning Agent as one of Oracle’s larger agentic experiences in development.
A 90-Day AI Readiness Roadmap and What Comes Next for Finance
For organizations asking how to begin preparing for AI, Mike points to three priorities.
1. Define the specific finance problem
Do not begin with, “Where can we use AI?”
Start with a finance problem worth solving.
Identify where users spend excessive time, where decisions are delayed, or where repetitive processes create friction.
2. Assess and align your data
Determine what information the AI needs and whether that data is actually ready.
That includes evaluating:
-
-
- Data quality
- Metadata
- ERP-to-EPM mappings
- Historical data
- Business rules
- Governance
-
Predictive use cases, for example, require enough historical information for the system to generate useful predictions.
3. Match the use case to the right tool
Organizations do not need to wait for fully autonomous finance.
Existing Oracle EPM capabilities can provide a practical starting point for learning how AI performs within real finance processes.
Start with built-in predictive analytics, then native assistants, then custom agents where gaps remain. Application-specific AI capabilities can create early wins while the broader AI roadmap develops.
Choosing The Right Implementation Partner
AI in EPM is only as good as the integration and metadata foundation underneath it, and that foundation is exactly where EPMI’s data integration and EDM expertise lives. EPMI helps organizations align ERP and EPM data, strengthen integration architecture, and identify where Oracle’s emerging AI capabilities can create positive impact.
FAQs
1. What is Enterprise Data Management(EDM), and why is it important for AI readiness?
Enterprise Data Management helps define and map metadata across ERP, EPM, and other systems, creating a more consistent source of truth and reducing the need for manual translation between platforms.
2. Why does chart of accounts alignment matter for AI in EPM?
AI-generated answers in EPM need to be traceable back to their source ERP data. Without aligned metadata, users may have difficulty validating or drilling through to those answers, which can reduce trust in the output.
3. Is autonomous finance available today?
Not at full scale. Most organizations are still in an assistant phase, where AI supports specific tasks. Agentic collaboration and greater autonomy are the next stages.
4. Should we build a custom AI agent or use Oracle’s native agents?
Start with the business problem. If an existing Oracle agent, assistant, or analytics capability already addresses the need, use that first. Custom builds in AI Agent Studio make more sense when a genuine gap remains.
5. Does this apply if we use a non-Oracle ERP with Oracle EPM?
Yes. Organizations using SAP or other third-party ERPs can still use EDM to align metadata or configure mappings directly within EPM.
Watch the AI World Series on YouTube!