AI Adoption in Finance
Monday, 16 February 2026
For two years, every EPM conversation eventually came down to one question: Is AI actually ready for finance work, or still a roadmap slide? That question is settled. It's time to move on to a harder one.
The adoption debate is over, but finance is still behind
Across industries, finance functions have roughly doubled their AI usage in the past year, and forecasters expect nearly all finance teams to be running at least one AI-enabled tool within the next 12 months. Eighteen months ago, the honest pitch to a CFO was "here's what AI could eventually do for your close, your forecast, your consolidation." Today, most already have the toggle sitting inside the platform they already pay for.
But adoption numbers hide an uncomfortable truth: finance remains one of the slowest-moving functions in the business when it comes to AI, trailing well behind engineering, marketing, and customer service. Most teams that have "adopted" AI are what one recent industry report bluntly calls tinkerers — experimenting with a chatbot for research or memo-writing, copying outputs in and out by hand, with no governed data underneath. Few have genuinely embedded AI into core planning and reporting.
That gap is the real story for 2026. The risk isn't a finance team failing to adopt AI — it's adopting it shallowly, at the edges, while the deeper capabilities already sitting in their Oracle EPM, Board or Jedox environment go untouched.
Why "AI-powered" isn't a pitch anymore
Every EPM implementation partner now sells some version of AI activation in your existing environment. The best-of-breed vendors we work with have made that easy — arguably too easy. Predictive forecasting, anomaly detection, generative narrative drafting, planning agents: these are vendor-shipped capabilities now, not bespoke builds. Flipping them on is configuration, not differentiation.
That's a good thing for the technology, but it means a consulting partner's value can no longer be "we can turn AI on for you." It has to be harder to copy: knowing which capabilities are trustworthy enough for an audit committee, which need a human in the loop, and where the model's blind spots sit inside a specific business's data.
That last part plays out differently by industry — which is the point. In retail, an anomaly-detection model watching stock coverage is only as good as the promotional calendar feeding it. In a life insurer, the same category of capability sits at the seam between actuarial assumptions and financial reporting, where an unexplained reserving variance isn't noise — it's a pricing risk that surfaces as an audit finding six months later. Same platform capability, completely different judgment required to trust it.
Where the AI focus in EPM needs to go next
- Governance before generation. Fewer than one in five organisations have anything like a comprehensive AI governance framework, even though governance is what separates teams getting real financial impact from those just running pilots. Any AI capability going into planning or reporting needs an answer to "how would we explain this to an auditor?" before "how much time does this save?"
- Value from what you already own, not a re-platform. The platforms have shipped capability faster than most organisations have absorbed it. Most finance teams are still underusing AI features already built into tools they're already paying for. The immediate opportunity isn't a new AI initiative — it's finally configuring and trusting the AI-powered tools already in Oracle EPM, Jedox or Board, rather than waiting for the next major transformation program.
- A human who owns the exception, not just the model. The most useful AI feature isn't the one that generates the answer — it's the one that reliably flags when the answer looks wrong. That only works if someone with real domain judgment is genuinely positioned to act on the flag, not rubber-stamp it. Weak data foundations remain the biggest obstacle, well ahead of any shortage of AI tooling.
The AI conversation in EPM has moved past the 'whether' stage. The organisations that get real value from it in 2026 will be the ones that treat "governed and understood" as the bar, not "switched on" — whatever industry they're reporting for.
