What Two EPM Veterans Think About AI
August 25, 2026 by Tim Faitsch
I recently sat down with Kyle Goodfriend, Principal Solutions Engineer at Oracle, and Peter Nitschke, Founding Director of Pivot2 Solutions in Australia, to talk about where AI is really landing in the EPM world. The conversation ran long because there was a lot to talk about. We barely scratched the surface but I want to cover what stood out to me.
The platform’s AI features aren’t where the value is yet.
Oracle markets AI heavily inside EPM: Insights, Predictive Planning, Advanced Predictions. Kyle’s finding something different in the field. The bigger wins are coming from using AI to build tools that operate on the platform, not from using the AI features baked into the platform itself.
His example was that customers doing Advanced Predictions almost always hit the same wall. They don’t know which accounts correlate with the ones they’re trying to predict, and they don’t know which drivers to use. That’s usually where the project stalls.
Kyle tried something different with a struggling customer. He asked an AI coding assistant to write a Groovy rule that would scan every balance sheet account, test each one for correlation against five target accounts, check for lagging indicators up to six months out, and spit out a confidence score for each result into a CSV report. He’d never have attempted that by hand. Forty minutes into the call, they had a working correlation report. By the end of the session, they had a full multivariate prediction model with lagging indicators built in.
Kyle put it bluntly. He doesn’t write Groovy anymore. (This shocked me since he’s probably the most well-known Groovy expert in the EPM world.) Now he tells the AI what he wants, and it writes the Groovy for him. What used to be a skill gap between “I know what I want” and “I know how to code it” has basically collapsed. That’s a bigger shift for EPM implementations than any dashboard feature.
Peter had a version of the same story from the other side of the world. One of his functional analysts, someone who has never written code, used an AI tool to build a one-page balance sheet report that looked better than anything the team had produced manually. The analyst then sent over an Excel file mapping out the measures and formulas needed to load it. Peter’s read was that the code looked right. That kind of output, from someone with zero development background, seems to be the new norm.
The bottleneck is us.
Kyle made a point that’s stuck with me since the call. For his entire career, the limiting factor in what technology could do was always hardware: processor speed, graphics capability, storage. That’s not true anymore. As he put it, the limitation is now the human.
He gave an example from banking where an AI system can not only flag that you missed your loan targets, it can explain why (approval times too slow, credit thresholds too tight, etc) and recommend specific fixes like shifting marketing spend to a particular region or adjusting risk tolerance. That capability exists in the Oracle EPM platform today. But are people ready to trust and act on what the system tells them?
People think AI is magic, not math.
Kyle admitted this was his own blind spot too. He built a forecasting application and told it to act like a mortgage banker predicting loan volume. When he asked how it arrived at its numbers, he expected something closer to intuition. Instead, it told him plainly: it used statistical methodology on historical data. Math.
That distinction matters more than it sounds. Peter’s take was that a lot of resistance to AI in finance comes from people treating it as a black box that “just makes stuff up,” when in reality it’s applying identifiable statistical models. Once people understand it’s not guessing, and that it can show its work, trust tends to follow.
But Peter also flagged the flip side. Ask an AI tool for a forecast, start a fresh session, ask again, and you’ll get a directionally similar but numerically different answer each time. Most people don’t have strong intuitions about statistics and probability, and that variability can undermine confidence just as easily as it can build it. His concern for planning and forecasting specifically: organizations ending up with five different AI-generated forecasts, each defensible, each polished, none of them able to actually anchor a decision. A forecast’s job is to guide the next decision. Five versions of a forecast just muddies the water.
This proliferation of content was something I wish I could have covered during the call. It’s so easy to generate content with AI whether that be long-winded emails or more forecasts than anyone has time to review.
Budgeting won’t disappear.
Peter had been on record predicting that COVID would kill the annual budget cycle. Everyone’s numbers got blown up mid-year, and he figured no one would bother with rigid annual budgets again. Three years later, most organizations are still doing exactly that.
His read now is that AI won’t kill budgeting either, and not because it can’t generate the numbers. Budgeting was never really a quest for the most accurate number. It’s a quest for a number that someone in the organization is willing to own. A junior analyst can produce the numbers, but they can’t own the budget. If you offload too much of the thinking, whether to AI or to a junior analyst, you lose the ownership that makes a budget mean anything.
Where he does see real change is in fast, disposable micro-forecasts. Quick, throwaway models for sales, ops, inventory, workforce, or retention that give you directional guidance on what to do now versus what to revisit in three months, without the rigid infrastructure most forecasting tools carry.
The closing thought.
Kyle ended with an analogy that captures the tension in the whole conversation. Using AI without any training is like handing a six-year-old the keys to the car. He said if he had to choose between spending his budget on a laptop or on AI, he’d skip the laptop. What he gets out of AI matters more to his productivity than the machine it runs on.
That’s really the throughline from this discussion. The tools are already capable of more than most organizations are using them for. The gap right now is education, trust, and a willingness to actually learn how these systems work instead of treating them as either magic or a threat.
Related Video: Listen in on this intriguing discussion here.