Everyone's Using AI in Finance. No One Knows What They're Doing Yet.

# Everyone's Using AI in Finance. No One Knows What They're Doing Yet. Gartner reports that 59% of finance leaders are already using AI. Two-thirds of CFOs call it "critical to performance." If you believe the vendor pitches, AI is revolutionising accounting by automating data entry, invoice reconciliation, and expense reporting. Which is to say: we've invented expensive software to do what good accounting software already did. ## The Pattern Recognition Problem The current wave of AI implementation in finance reads like a greatest hits of tasks we've been automating for decades. Invoice processing. Expense categorisation. Bank reconciliation. These aren't new problems, and AI isn't solving them in fundamentally new ways. It's just doing pattern matching faster and with more convincing marketing copy. Where AI does show promise is in the stuff accountants actually hate: hunting through unstructured data, spotting anomalies in transaction patterns, and generating predictive insights from messy historical data. One US firm is using AI to forecast tax implications based on client decisions. That's genuinely useful, assuming the model doesn't hallucinate tax advice that lands everyone in front of the ATO. ## The Governance Gap Here's what's not making the promotional material: governance frameworks. As one report notes, finance leaders must establish standards that keep automation in check. Which sounds straightforward until you're explaining to the board why the AI flagged a legitimate supplier payment as fraud, or why it confidently reconciled accounts that don't balance. The CFOs making this work aren't the ones implementing "autonomous agents" across their finance function. They're the ones running small, contained pilots with clear success metrics and human oversight. They're asking: what specific problem does this solve, and what happens when it gets it wrong? ## Real-World Reality Check The finance teams seeing actual results aren't chasing "multi-step workflows" or "agentic AI" (whatever that means this quarter). They're targeting specific pain points: - Faster month-end close by automating journal entry validation - Better cash flow forecasting by pattern-matching payment behaviour - Reduced audit preparation time through automated documentation retrieval Notice what's missing? Revolutionary transformation. Paradigm shifts. The future of work. What's present? Measurable time savings on tedious tasks. Which is good! But it's not the fundamental reimagining of finance that the headlines promise. ## The Skills Question If AI is handling data entry and basic reconciliation, what are accountants supposed to do? The optimistic answer: higher-value analysis, strategic planning, and stakeholder management. The realistic answer: figure out why the AI is producing nonsense, fix the underlying data quality issues it keeps hitting, and explain to people why we can't just "let the AI handle it." Finance teams need people who understand both the accounting and the algorithms. Not because they need to code, but because someone needs to spot when the model is confidently wrong. Pattern-matching algorithms are brilliant until they encounter a pattern that's genuinely new. ## What Actually Matters If you're implementing AI in finance, three things matter more than the technology: 1. **Clear problem definition**: What specific task is taking too long or producing too many errors? If the answer is "everything," you're not ready. 2. **Human checkpoints**: Where in the process does a human need to verify the output? Hint: more places than your vendor suggests. 3. **Failure protocols**: What happens when it's wrong? How do you catch it, fix it, and prevent it happening again? The CFOs succeeding with AI aren't the ones betting the farm on autonomous finance functions. They're the ones running controlled experiments, measuring results, and treating AI like any other tool: useful for specific jobs, useless for others, and never a substitute for knowing what you're doing. ## The Bottom Line AI will change finance work. It's already changing it. But the transformation looks less like "revolutionary automation" and more like "faster, slightly smarter versions of things we already automated." Which is fine. Sometimes boring improvements are the best improvements. Just don't expect the AI to reconcile your accounts while you focus on strategy. Someone still needs to check its work, understand why it flagged that transaction, and explain to the auditors why you're trusting a black box with your financial data. That person is still you. The AI just means you can do it faster, until it doesn't.