How AI Agents Handle Month-End Close
What actually happens when you let Claude run the close process — and where humans still matter.
Month-end close is the accounting world's recurring nightmare. Every company does it. Nobody likes it. The process is a checklist of 30-80 tasks that need to happen in a specific order, with dependencies, handoffs, and the constant risk that one missed accrual throws off everything downstream.
Most accounting teams spend 5-10 business days on close. The industry has been promising "faster close" for decades. The tools have gotten better. The timelines haven't moved much.
AI agents change the math — not by making humans faster at the same tasks, but by handling entire categories of tasks autonomously. Here's what that looks like in practice.
What an AI Agent Actually Does During Close
An AI agent isn't a chatbot that answers questions about your data. It's a system that can execute multi-step financial workflows — reading data, applying rules, making decisions, posting entries, and flagging exceptions. Think of it as a junior accountant who never sleeps, never forgets a step, and processes transactions at machine speed.
During month-end close, an AI agent can handle:
Bank reconciliation. The agent pulls your bank statement, matches transactions against the GL using three passes (exact match, fuzzy match, pattern match), posts the matched entries, and surfaces the unmatched items for human review. A process that takes 2-4 hours manually takes minutes — and the exceptions list is usually 5-10 items instead of 50.
Accrual calculations. The agent looks at recurring expenses, open POs, and historical patterns to calculate and post accruals. It knows which vendors bill in arrears, which contracts have monthly true-ups, and which accruals from last month need to be reversed.
Revenue recognition. For companies on ASC 606 or IFRS 15, the agent evaluates contracts, identifies performance obligations, calculates the transaction price allocation, and posts the appropriate revenue entries. It handles the journal entries that most teams dread.
Intercompany eliminations. If you have multiple entities, the agent identifies intercompany transactions, verifies they balance, and posts the elimination entries. It catches mismatches before they become reconciliation problems.
Flux analysis. Once the books are tentatively closed, the agent compares the current period to prior period and budget, calculates variances, and drafts commentary for the significant ones. "Marketing spend increased 23% vs. prior month, driven by the Q1 campaign launch" — that kind of narrative, generated from the data.
Checklist management. The agent tracks which close tasks are done, which are blocked, and which are next. It knows the dependency chain — you can't do the flux analysis until the accruals are posted, and you can't post accruals until the bank rec is done.
Where Humans Still Matter
The AI handles volume. Humans handle judgment.
Exception review. The agent matches 90-95% of bank transactions automatically. The remaining 5-10% are the ones that need a human to look at — unusual amounts, new vendors, transactions that could go to multiple accounts. The AI surfaces these; the human decides.
Estimates and assumptions. Accruals for things like legal reserves, warranty provisions, or bonus pools require judgment about future outcomes. The AI can calculate based on your assumptions, but the assumptions themselves need a human.
Disclosure decisions. What goes in the footnotes, how to present a complex transaction, whether something is material — these are judgment calls with real consequences. AI can draft; humans approve.
Controls and sign-offs. Segregation of duties, management review, and audit trail integrity all require human accountability. The AI does the work; a human certifies it's right.
What a 2-Day Close Actually Looks Like
Here's a realistic close timeline with AI agents handling the execution:
Day 1:
- AI runs bank reconciliation for all accounts (automated, completed by 9am)
- AI posts standard accruals and reverses prior month (automated, completed by 10am)
- AI calculates and posts revenue recognition entries (automated, completed by 11am)
- AI runs intercompany matching and posts eliminations (automated, completed by noon)
- Human reviews all exception items and AI-posted entries (afternoon)
Day 2:
- AI runs trial balance and flux analysis (automated, completed by 9am)
- AI drafts variance commentary (automated, completed by 10am)
- Human reviews financial statements and variance commentary (morning)
- Human handles any adjustments, estimates, and judgment items (afternoon)
- Final review and close (end of day)
Two days. Not because the AI is magic — because the AI handled the 70% of close tasks that are rule-based and repetitive, leaving humans free to focus on the 30% that requires judgment.
The Skeptic's Questions
"How do I trust the AI's journal entries?"
Every entry posted by the AI has a full audit trail — the source data, the rule applied, the calculation, and the posting. You can review any entry the same way you'd review a junior accountant's work. The difference is the AI doesn't make typos, doesn't transpose digits, and doesn't forget to reverse last month's accrual.
"What about my auditors?"
Auditors care about controls, documentation, and accuracy. AI-posted entries have better documentation than most human-posted entries (every decision is logged). The control framework — review, approval, segregation of duties — still applies. The AI is the preparer; a human is the reviewer.
"What if the AI gets something wrong?"
It will. Just like humans get things wrong. The difference is that AI errors tend to be systematic (wrong rule applied to a category of transactions) rather than random (transposed a digit on one entry). Systematic errors are easier to find and fix because they follow patterns. Your review process catches them the same way it catches human errors — by looking at the output.
How Artifi Handles This
In Artifi, month-end close is a conversation with Claude. You tell Claude to run the close for March. Claude executes the task sequence — bank rec, accruals, revenue recognition, intercompany, flux analysis — and reports back with a summary: what was posted, what needs review, what's unusual.
You review the exceptions, approve the entries, and Claude closes the period. The entire interaction is logged, auditable, and reversible.
It's not a dashboard with a progress bar. It's a conversation with an agent that does the work and tells you what it found.
Related reading:
- Why Month-End Close Still Takes 10 Days — and How AI Agents Fix It
- From 10-Day Close to 2-Day Close: How Agent Orchestration Changes Everything
Artifi gives Claude the skills to run your month-end close. Learn more.