Day two of the close. The bank reconciliations are done, but receivables and payables are off from the general ledger by amounts nobody can explain yet, and the flux review will not start until they agree. The commentary for the management pack will be written late on day five by whoever is left.
Those are the three places where AI in the month-end close is most useful: reconciliations, flux analysis and the first draft of commentary. Each needs a different level of review, and treating them alike is where teams get into trouble.
How long a close takes, and where the days go
APQC measures the cycle time to complete monthly consolidated financial statements in calendar days, from trial balance to agreed consolidation. The cross-industry median is 6 days. Top performers sit at a median of 5 days, bottom performers at 10. The data was collected in 2022 and published on 5 May 2025, from a sample of 10,201.
Reconciliations are often where the days go. ISG (Ventana Research) reported in July 2022 that "59%" of companies completed the monthly close within six business days, against 60% in 2019. In the same research, 72% of firms that automated most or all reconciliations closed the quarter within six business days, compared with 25% of those with some or no automation. ISG did not state the sample, and an association is not proof that automation caused the faster close.
Shortening the close is mostly process work, covered in our piece on what a five-day close takes. AI helps most once that process exists.
Reconciliations: matching is the reliable part
Matching bank lines, subledger balances and intercompany entries is high-volume and rules-heavy, which suits software. Microsoft's prerelease Financial Reconciliation agent for Excel (documentation updated May 2026) groups transactions as perfectly matched, potentially matched or unmatched. In Dynamics 365 Finance, account reconciliation (documentation updated 1 September 2026) compares the general ledger with the payables, receivables, tax and bank subledgers, and its Copilot-powered agent, where enabled, proposes the most likely action with an analysis and a justification.
Perfect matches can usually be accepted once sample testing supports it. The reviewer's time belongs on the potential matches, the unmatched items and any proposed adjustment. Intercompany is harder again, as we discussed in July's piece on intercompany.
Flux analysis: calculating is easy, explaining is not
Flux analysis compares each balance with the prior month, the budget or the same month last year, and asks why it moved. AI does the first half well: it calculates every movement, applies the agreed thresholds and lists the journal lines behind each one over the limit.
It can also flag what is missing. A recurring accrual booked in each of the last twelve months and not in this one produces no transaction to explain, so a review based only on movements can miss it. Comparing this month's entries with the usual pattern catches that.
Explaining the cause is where review is needed. Take legal fees that doubled in a month. A model can see the increase and that most of it came from one vendor. Whether that is a one-off dispute, a new matter that will run all year, or an invoice in the wrong period, it cannot know unless someone tells it.
Draft commentary: a first draft only
With the movements and transactions in hand, AI can write a readable first draft. Figures should come from the ledger, not the model, and tie to the trial balance. The reviewer checks cause, emphasis and omissions.
| Close task | What AI does reliably | What needs review | Control |
|---|---|---|---|
| Bank and subledger reconciliations | Matches transactions and sorts them by confidence | Potential matches, unmatched items, proposed adjustments | Preparer and reviewer sign-off unchanged |
| Ledger-to-subledger differences | Suggests a likely action with an analysis | Any journal it implies | Normal journal approval limits |
| Flux analysis | Calculates movements, applies thresholds, lists transactions, flags missing recurring entries | Whether the stated cause is right | Thresholds documented and approved |
| Commentary | Drafts text from the flux results | Cause, emphasis and omissions | Named reviewer; figures tied to the ledger |
Controls for AI in the month-end close
Separate suggestion from posting. AI proposes; journals follow the approval route you already have. Read-only access covers everything above.
Keep a named preparer and reviewer. Whoever accepts an AI suggestion becomes its preparer, and the reviewer's sign-off means what it meant before.
Log the proposal and the decision. Record what the tool suggested, what was accepted or changed, and by whom. Auditors will ask for it.
Run it in parallel first. For the first few closes, run AI matching alongside the existing reconciliation and compare. Rely on it only where the comparison holds.
Tell the auditors early. Agree how they will test an AI-assisted reconciliation before the year-end, not during it. We cover the wider control questions in AI Internal Controls: What Changes When AI Touches Financial Reporting.
What AI in the month-end close will not fix
It will not fix late inputs. If supplier invoices, payroll files or operational data arrive on day four, the close finishes after day four, however fast the matching runs.
It will not fix balances nobody understands. Matching on top of an account that has not been properly reconciled in a year produces a tidy view of an unknown number. Clean up the opening position first, then automate.
Frequently asked questions
How is AI used in the month-end close?
AI is used mainly to match transactions in reconciliations, calculate and explain movements in flux analysis, and draft first-version commentary. People review the exceptions, the causes and anything that leaves the finance team.
Can AI do account reconciliations?
AI can reliably do the matching part, sorting items into matched, potentially matched and unmatched. Potential matches, unmatched items and proposed adjustments still need a preparer and a reviewer.
What is flux analysis, and can AI do it?
Flux analysis compares balances with a prior period or budget and explains material movements. AI can calculate the movements and gather the transactions behind them, but someone who knows the business must confirm the cause.
Does AI-written close commentary need review?
Yes. Figures should come from the ledger and tie to the trial balance, and a named reviewer should check causes, emphasis and omissions before the commentary is used.
How long should a month-end close take?
About six calendar days is typical: APQC's cross-industry median from trial balance to agreed consolidation is 6 days, against 5 for top performers and 10 for bottom performers (2022 data, published 2025).
Where CFOLogic fits
We provide accounting operations, the close included, and we design AI-assisted close tools that sit on the client's existing systems and read from them, so that nothing reaches the ledger or leaves finance without a person approving it. A first proof of concept would normally cover one part of the close, such as subledger reconciliations or flux commentary. See how we run finance operations for growing companies.