Why the Agentic EPM Era Demands Foundational Context

As Oracle EPM AI features command the spotlight, we face a critical question: what should we actually deploy, and what is merely noise? To answer that, we have to look past the current hype and look back at how we got here.

From database science to the SaaS abstraction

I have been deploying Oracle EPM solutions for well over 20 years. That would mean before Planning and Essbase became Oracle offerings; they were Hyperion offerings. At the time when I was learning the trade of Essbase, it was important to understand the basics of what a multi-dimensional data model is, what index and page files are, and given the limited compute, what an optimal multi-dimensional Essbase database design should be. It was important to ask these questions to build and deliver a scalable solution for clients that lived up to their demands of performance reporting, followed by performance planning against the actual performance reporting. At my best, I had become a Database Consultant perpetuated by the learnings of the data science behind the multi-dimensional data model in Essbase.

Then in 2015, I deployed my first Oracle Planning and Budgeting Cloud Service (PBCS) application for a Real Estate Construction and Development client. It was new, exciting and in retrospect a watershed moment that meant a lot of my Essbase fine tuning and optimization skills at the database level were now being automatically handled by Oracle Cloud, now a SaaS offering. Creative autonomy still existed in terms of building interfaces, moving data, writing calculations and business rules, writing reporting and dashboard insights et al, but I did not have to worry much about right-sizing the Essbase database — well, not that much anyway. I was becoming an Application Consultant perpetuated and supported by my knowledge as a Database Consultant.

I witnessed the next generation of Planning and Essbase professionals coming through the ranks in consulting, and they all began to enter as Application Consultants building solutions without a care in the world for the database aspects of Essbase. I vividly remember sitting down with a Senior Consultant who was trying to emulate a Thick Ledger in Essbase to model forecast data — future ERP was their forecasting solution — and had blown past the 2^52 limit for sparse dimension intersections, then blamed the Oracle Cloud instance for the critical error they were hitting. What I remember most is sitting that consultant down, explaining the limits (by the way, that is quite a limit) and getting down to key Essbase multi-dimensional science.

All this information was available ubiquitously on the web and was there for the taking. So knowledge was available, the means to access and learn from it were available, and the consultant by all means was a hard worker with the willingness to learn. But what was missing was the context of why Essbase's scientific limit is important to consider. This gap, I came to believe, was the side effect of the SaaS revolution which was starting people as Application Consultants at the very beginning of their careers.

A new march is afoot

Another march is afoot as I write this article. I see a new slate of professionals entering the Oracle EPM space who are aspiring to build Oracle EPM Agents. With basic application module understanding and a grounding functional knowledge of Financial Planning and Analysis, these consultants are proposing Agentic solutions to the marketplace. They are aspiring to make the next leap to Agentic EPM Consultants. Again, at this time, all information is available; the tools and means for building EPM agents are becoming pre-eminent through AI Agent Studio, and access is completely democratic.

Just as SaaS abstracted away the database layer, Generative AI threatens to abstract away the application layer entirely. The new trap is believing that a well-prompted agent can replace a robust data strategy.

The failure you won't see coming

An LLM doesn't inherently understand structural data integrity or the strict guardrails of enterprise financial models. If the new generation of consultants builds agents without understanding the underlying EPM architecture, the failure won't be a visible 2^52 limit error. It will be an agent confidently generating a flawed forecast that a leadership team acts upon. So what do we deploy? We do not deploy AI features simply because they are available. We deploy them with the same structural rigor that we used to design optimal Essbase databases 20 years ago. The value of an EPM consultant is no longer in writing the code or configuring the widget — it is in providing the context.

True intelligence in enterprise EPM isn't the autonomous agent spinning up a forecast; it is the human practitioner who ensures the agent is grounded in systemic truth. As we enter this agentic frontier, the challenge for our profession is clear: we must ensure that our leap into AI is supported by the bedrock of data science, not just the illusion of capability. The agents are coming — but they are only as good as the context we give them.

Looking ahead: putting theory into practice

Context is our north star, but how does it manifest in the real world? Over the next few articles, we will move from philosophy to the field. I will be opening the hood on the current landscape of Oracle EPM AI-driven solutions to stress-test what is actually ready for enterprise deployment. We will break down and investigate the practical realities of two distinct paths.

The first is the Embedded Capabilities: the native, out-of-the-box AI features built directly into the Oracle EPM ecosystem. The second is the Agentic Frontier: custom, at-times autonomous workflows built using frameworks like AI Agent Studio. We will evaluate where these tools provide genuine architectural leverage — and where they risk tripping over the same foundational limits we've been navigating for two decades. Stay tuned.