When AI runs the numbers: what CFOs must actually do differently

A mid-sized European manufacturer reduced its monthly close from 12 days to 4 days in 2025. The tool responsible was not a custom-built enterprise system costing millions. It was a combination of Microsoft Copilot integrated into their existing ERP and a modest internal data governancedata governanceData governance is the set of policies, roles, and processes that ensure data is accurate, secure, well-defined, and used responsibly across an organization.View full definition → effort that had been underway for 18 months prior. The CFO's comment afterward was instructive: "The AI was the easy part. Getting our chart of accounts consistent across subsidiaries was the hard part."

That comment cuts to the center of what is actually happening in digital finance right now. The technology is maturing faster than most finance organizations are prepared for, and the gap is not primarily a technology gap.

The state of AI in finance functions in 2026

The trajectory has become clear over the past two years. Large language models, predictive analytics, and process automation have moved from finance team experiments into production environments at companies of meaningful scale. JPMorgan Chase has been running its COiN contract intelligence platform for several years now, processing documents in seconds that previously required thousands of hours of legal review. BlackRock's Aladdin platform continues to expand its AI-driven risk modeling capabilities across both internal and client portfolios. These are not edge cases.

What has shifted more recently is the penetration into mid-market finance functions. Platforms like Oracle Fusion, SAP S/4HANA, and Workday have embedded generative AI features directly into FP&A workflows, variance analysisvariance analysisVariance analysis compares actual financial results against budgeted or planned figures to quantify differences and explain why they occurred.View full definition →, and cash flow forecasting. A finance team that wants AI-assisted scenario modeling in 2026 does not need to build anything from scratch. They need to decide whether to use what their existing vendors are already offering.

That last point deserves scrutiny. Vendor-reported productivity gains should be read carefully. When SAP or Workday publish figures showing 40% reductions in reporting cycle time, these are not independent audit results. They come from curated case studies with favorable conditions. Independent research from Gartner and Accenture's Institute for High Performance paints a more measured picture: meaningful productivity gains are real, but they concentrate in organizations that had already invested in data qualitydata qualityThe degree to which data is fit for purpose: accurate, complete, consistent, timely, valid and unique. Poor quality data undermines analytics, reporting and AI.View full definition → and process standardization before AI was layered on top.

The other significant development is agentic AIagentic AIAgentic AI refers to AI systems that pursue goals autonomously by planning, taking actions through tools, and adapting based on results, with minimal step-by-step human direction.View full definition →: systems that do not just answer questions but execute sequences of tasks autonomously. In finance, early deployments involve accounts payable matching, anomaly flagging in expense reports, and treasury cash positioningpositioningThe mental space you want your brand to occupy in your target customer's mind relative to alternatives.View full definition →. The productivity case is compelling. The control and audit case is more complicated.

What this means for the CFO

The CFO's role is shifting in a specific and uncomfortable direction. For the past decade, the story was about moving finance from scorekeeper to strategic partner. AI is now doing something more disruptive: it is compressing the time required for analytical work to the point where the question is no longer "how do we get the analysis done faster" but "what do we actually do with the judgment time that has been freed up?"

That is a harder question than it sounds, because most finance organizations are not structured to answer it. Headcount reductions may follow, but the CFO who leads only with cost cutting will miss the more important opportunity. The teams that perform best over the next three to five years will be the ones that redirect analytical capacity toward decisions that genuinely required human judgment all along but rarely received it. Acquisition modeling with real uncertainty ranges. Capital allocation debates that account for second-order effects. Stress testing assumptions that no one previously had time to stress test.

There is also a control dimension that is not optional. When an AI system flags a cash flow anomaly or proposes a journal entry adjustment, who is accountable for the decision? Finance leaders deploying agentic tools without clear human review protocols are creating audit exposure they may not fully see yet. The PCAOB and several European regulators have begun issuing guidance on AI-assisted audit processes, and the direction of travel is toward requiring documented human oversight at defined decision points, not blanket prohibition, but traceability.

Talent is the third pressure point. The finance professional who will be most valuable in 2028 is not the one who can build a three-statement model in Excel from scratch in two hours. That skill will be table stakes or obsolete. The valuable profile is someone who can interpret AI outputs critically, identify where the model's assumptions do not fit the business context, and communicate uncertainty clearly to non-finance stakeholders. CFOs who are not explicitly developing this capability in their teams are building a fragile function.

Practical priorities for CFOs moving forward

• Audit your data infrastructure before expanding AI tooling. The most common reason AI finance deployments underperform is not the algorithm, it is inconsistent data definitions across systems. Fix the foundations first.
• Establish a clear policy on AI-generated outputs entering the financial record. This means deciding which outputs require human sign-off, at what materiality threshold, and by whom. Write it down before you need it.
• Push your Big Four auditor on their own AI approach. Firms like Deloitte and EY are deploying AI heavily in audit procedures. Understanding how that affects your audit trail, sampling methodology, and documentation requirements is now part of managing the audit relationship.
• Distinguish between AI that assists decisions and AI that makes decisions. The former is largely a productivity tool. The latter requires governance architecture. Many finance teams are running the second type while telling themselves it is the first.
• Build at least one internal case study before 2027. A real, documented example where AI changed a financial outcome, with specifics on what worked and what did not, gives you both credibility with the board and a learning baseline for the next deployment.

The CFO who will have the most influence over the next five years is not the one with the most sophisticated AI stack. It is the one whose organization can be trusted to produce financial intelligence that boards and investors believe. AI makes speed easier. It does not automatically produce trustworthiness. That part remains a human responsibility, and it is the CFO's responsibility specifically.