Generative AI has moved from proof-of-concept to active deployment in finance functions faster than most technology adoption cycles. CFOs who were cautiously experimenting in 2023 are now running live pilots on month-end close, variance analysis, and disclosure drafting. The question is no longer whether AI will change financial reporting — it is how to deploy it safely, at what pace, and with what governance.
Where AI Is Adding Real Value Today
The most mature AI use cases in finance fall into three categories:
- Transaction processing — automated GL coding, invoice matching, and bank reconciliation using classification models trained on historical patterns. Mature technology with high accuracy rates in structured environments.
- Anomaly detection — AI models that flag unusual journal entries, outlier transactions, and control exceptions for human review. Increasingly used as a first line in financial controls and audit preparation.
- Narrative generation — large language models (LLMs) that produce first drafts of management commentary, variance explanations, and board reports from structured financial data inputs. Requires careful human review but significantly accelerates drafting cycles.
The Disclosure Drafting Use Case: Promising but Requires Caution
Generating first drafts of financial statement notes and management commentary is one of the most discussed AI applications in finance. The appeal is clear: disclosure drafting is time-consuming, repetitive in structure, and often done under year-end time pressure. LLMs can produce technically plausible drafts in seconds.
However, several risks demand careful governance:
- Hallucination — LLMs can generate plausible-sounding but factually incorrect statements. Financial disclosures require factual accuracy against source data, not linguistic plausibility.
- Standard currency — AI models trained on historical data may not reflect recent standard changes (e.g., IFRS 18, new ESRS requirements), producing disclosures that are structurally outdated.
- Audit trail — AI-generated disclosures must still have a traceable basis. Auditors will ask where each number and assertion comes from.
"AI accelerates the drafting process — but the finance team remains fully accountable for every word in the financial statements. Governance frameworks must reflect that accountability."
What Auditors Are Doing About AI
The major audit firms are actively integrating AI into audit processes — using it for data analytics, sample selection, and document review. This has implications for finance teams: auditors using AI-assisted review may ask different questions, spot different patterns, and focus on different risk areas than a purely manual audit. Finance teams should expect AI-augmented audit procedures to evolve rapidly over the next two to three reporting cycles.
Audit standard setters are also responding. The IAASB has issued guidance on the use of technology in audits, and regulators are beginning to ask how AI tools used in preparing financial statements are governed and documented.
Building an AI-Ready Finance Function
Finance teams that want to deploy AI responsibly should focus on four foundations:
- Data quality first — AI amplifies data quality problems. Clean, well-structured financial data is a prerequisite for reliable AI outputs.
- Human-in-the-loop design — all AI outputs that feed into financial statements or external reports must have a defined human review and approval step.
- Documented use policy — establish which AI tools are approved for use, what they can be used for, and what outputs require which level of review.
- Audit committee briefing — boards and audit committees should understand where and how AI is being used in the financial reporting process before external auditors ask.
The Near-Term Horizon
Over the next 18 months, the most significant AI developments for finance teams are likely to be in real-time close processes (continuous accounting), intelligent XBRL tagging, and AI-assisted CSRD and sustainability data collection. Finance leaders who build AI governance frameworks now — rather than retrofitting them after deployment — will be better positioned to scale responsibly as the technology matures.