AI is coming for the month-end close - what CFOs should automate first
Thu, 23rd Jul 2026 (Today)
Many finance leaders have now sat through at least one demo promising an autonomous close, where software handles the reconciliations and the variance questions and the board pack is ready before the team logs on.
The reality inside most Australian finance functions looks a little different. Consero's 2026 CFO Survey put AI adoption in finance departments at 97%, effectively universal among the investor-backed companies it benchmarks. The depth of that usage is another question: a General Atlantic poll found only 17% of finance teams using AI in their core workflows, with 45% stuck in limited pilots and finance ranking last among business functions for workflow automation.
Most of that gap comes down to starting in the wrong place. AI handles close work well when the task is repetitive and the output is easy to check. It handles judgment calls badly, and nobody can hold a model accountable for a decision it gets wrong. CFOs who separate the two will close faster without losing control.
Bank and transaction reconciliation is the most AI-ready work in the entire close. The task is repetitive and the data is structured. A wrong suggestion is also cheap to catch, because a human reviews the exceptions anyway.
The major ERP vendors are converging on this. NetSuite's latest release is a useful barometer of where the category is heading: it adds an AI transaction matching assistant that recommends a single likely general ledger match when bank reconciliation throws up multiple candidates, with a rationale attached so the accountant can see why the suggestion was made. The rationale is an important inclusion because finance AI that cannot show its working tends to stall at audit, whatever it saved during the month.
For most mid-market finance teams, matching and reconciliation is where AI delivers its first measurable time saving, often within a single close cycle. If your team is still clearing bank lines manually, this is the place to begin.
The second candidate is the work nobody budgets time for. Close managers spend a disproportionate share of the cycle chasing missing accruals and querying balances that moved for reasons nobody can immediately explain.
AI suits this work because it is pattern surveillance across thousands of transactions. Modern close management tools flag exceptions and surface expected activity that has not appeared, so issues emerge on day two rather than day eight. The newer capability is automated flux analysis, which drafts an explanation of why account balances changed between periods. NetSuite is shipping a version of this in the same release and comparable functionality is appearing across the financial software market.
The advisory point for CFOs is to treat these tools as a first-pass analyst. The AI drafts the variance commentary and your controller still owns it.
Right now, AI is demonstrably good at one slice of forecasting: prediction from historical behaviour. Payment date prediction, which estimates when a customer will actually pay an invoice based on past behaviour, is now embedded in mainstream ERP banking modules and gives treasury teams a sharper view of near-term cash.
However, the hype still outruns the evidence in strategic forecasting. Revenue projections that hinge on pipeline judgment or a pricing call still belong to humans. The model only sees historical data, which means AI can draft the forecast, but you will need people who know the business to say which numbers to believe.
Other capabilities are further off again. Fully autonomous journal posting without review invites errors you will not find until audit. AI-drafted disclosures and board commentary produced without controller sign-off create accountability gaps that ASIC and your audit committee will not wear. And any tool that cannot show its working should not touch your general ledger, whatever the demo looked like.
Automate first where volume is high and the rules are stable, with a human already checking the output as part of the existing process. Reconciliation and matching meet that test today, and exception detection meets it with light supervision. Judgement work does not.
Before any of it, get your master data in order, because AI trained on inconsistent coding will automate your existing mess at speed. Governance needs settling up front too, deciding which AI features are enabled and who reviews the output. The vendors are making this easier, with admin-level AI controls now standard in major ERP releases.
None of this makes the close autonomous, this year or next. It can take days out of the cycle, and the teams already seeing that started with reconciliation, the least glamorous line on the AI roadmap.