93% of CFOs Use AI in Finance, But Human Judgment Remains Critical for Trust and Accuracy

AI Adoption in Finance Reaches Critical Mass

AI in finance has shifted from experimental projects to core operational infrastructure. According to Deloitte's CFO Signals, close to 93% of CFOs now use AI tools across key finance operations, marking a dramatic transformation from tentative pilots just two years ago1

1

. This AI adoption in finance reflects a fundamental shift in how financial teams approach automation, moving beyond simple task completion toward strategic integration. Yet the organizations achieving meaningful results aren't automating indiscriminately—they're making deliberate choices about where AI-powered finance adds genuine value.

The Trust Gap in AI Tools

Despite widespread AI adoption in finance, a significant trust deficit persists. Less than half—46%—of users trust what LLMs tell them, according to recent studies2

2

. More concerning, two-thirds of users admit they rely on AI output without verifying accuracy, and over half report making workplace mistakes because of it. This paradox reveals a critical vulnerability in AI for financial reporting: users simultaneously distrust and depend on these systems. The issue intensifies when financial modeling involves investor presentations or board reports, where AI hallucinations can undermine credibility. One of the Big Four accounting firms fell victim to publishing AI-generated data hallucinations, demonstrating that even sophisticated organizations aren't immune to these risks of AI in finance2

2

.

Human Judgment in AI: Relocating, Not Replacing

The evolution of AI-powered finance isn't eliminating human judgment—it's repositioning where that judgment applies. Finance teams face scenarios where data alone fails to capture organizational context. A spike in supplier costs might resemble a contract issue until you understand it reflects a strategic shift agreed in leadership discussions. A variance might trigger alarms until you recognize it's tied to planned operational changes1

1

. AI tools can highlight anomalies, but they cannot intuit the business nuance behind them. FTI Consulting's Global CFO Survey found that nearly 70% of CFOs now rely on AI-driven insights alongside experience-based judgment rather than as a replacement1

1

. This approach positions human judgment in AI at the strategic level—setting parameters, deciding what requires attention, and intervening when patterns don't align with business reality.

Data Quality: The Real Bottleneck

The biggest obstacle to effective AI in finance isn't model sophistication—it's data quality underneath those models. Finance teams have spent years connecting accounting systems, expense platforms, procurement tools, and spreadsheets. When these systems contain duplicate suppliers, inconsistent coding, or outdated information, AI doesn't fix the problem; it accelerates the consequences1

1

. Clean, consistent, and accessible financial data provides AI with reliable foundations. Fragmented data makes even sophisticated automation harder to trust. The most critical AI work happens before deployment: connecting systems, cleaning data, and establishing a single accurate view of financial information. Organizations making progress solve specific problems well—speeding up invoice processing, tightening expense categorization, spotting spend anomalies early—then scale once they understand how systems behave within their actual workflows.

Financial Modeling and Investor Scrutiny

When seeking funding or investment, the limitations of AI for financial reporting become stark. Prompting an LLM for three-statement financial models can generate professional-looking outputs, but the cookie-cutter perfection fades under scrutiny. AI models often rely on underlying assumptions too general to reflect specific business realities. Revenue projections, margin calculations, and EBITDA figures unravel under basic investor due diligence2

2

. LLMs don't know the terms of your last investment round, which customers might churn, how long deals actually take to close, or what your next product launch will do to margins—exactly what investors will test. This gap between plausible-looking financial modeling and business-specific accuracy underscores why experienced CFOs remain essential for defending numbers when funding rounds hinge on margin defensibility.

Data Exposure and Governance Risks

Beyond accuracy concerns, feeding unredacted cash-flow forecasts, margin structures, supplier terms, or runway projections into publicly available LLMs creates serious financial governance and compliance liabilities. Without proper guardrails, organizations risk exposing competitive intelligence while inadvertently training models that competitors might access2

2

. This data exposure represents a blind spot many overlook in their rush toward automation. Strategic accountants and CFOs who utilize automation tools treat them as junior analysts that accelerate scenario analysis and stress-test exit strategies, but they layer business and market knowledge that LLMs cannot yet consider. Decision-making around sensitive financial information still requires human oversight to maintain transparency and protect proprietary data.

Building Trust Through Strategic Integration

Trust in AI-powered finance shouldn't be an afterthought added through policy documents or training sessions. It needs embedding into how these systems function. Finance teams require visibility into how outputs are reached to challenge them when necessary, alongside clear boundaries around what AI systems can do independently versus when they should return decisions to human judgment1

1

. An AI tool that drafts reports differs fundamentally from an agent that recommends suppliers, flags unusual spending, or initiates payments. As AI takes on more responsibility, the question shifts from whether it can complete tasks to whether finance teams genuinely trust the results when small errors carry real financial or regulatory consequences. The real opportunity isn't creating finance functions that do less—it's enabling teams to spend less time on administration and more on decisions that matter, with human judgment in AI ensuring accountability at critical junctures.