Month-end close: what the agent prepares

Most of the close is preparation, and that is what the agent does: reconciliations, accrual and prepayment schedules, revaluation, intercompany matching, flux analysis and the checklist itself. It drafts and parks; it posts only certified recurring entries. The controller reviews every judgement, approves the journals that need it and signs the close. Nobody but a person locks the period.

Who owns the close, and what changes with an agent?

The month-end close turns a month of transactions into a set of books someone will sign. The financial controller owns it: the calendar, the checklist, the cut-offs and the final review. Accountants own their areas, such as payables, fixed assets or revenue, and prepare the entries and reconciliations for them.

With an agent, the accountants' first drafts arrive earlier and more complete. Their role shifts from preparing to reviewing. The controller's role does not change: the close is still signed by a person who can explain every balance. Whether an agent can run the month-end close by itself has a clear answer: no, and this playbook shows why it does not need to.

What does the agent prepare, day by day?

Close calendars vary. This one assumes a five-day close.

1. Day minus one, agent, depth 1: pre-close checks. Unposted invoices, unapproved journals, open purchase orders with receipts, suspense balances, and subledger-to-ledger differences. The list goes to each owner before the month ends.
2. Day one, agent, depth 2: accruals. Goods received not invoiced, recurring services without invoices, payroll accruals. Each is parked with its basis; see running month-end accruals with an AI agent.
3. Day one, agent, depth 3 once certified: recurring template journals. Prepayment releases, depreciation from the register, standing allocations. Amounts from sealed schedules only.
4. Day two, agent, depth 2: revaluation and intercompany. Foreign currency revaluation parked for approval; intercompany balances matched and differences listed by counterparty.
5. Day two to three, agent, depth 2: balance sheet reconciliations. Prepared with support and explained differences, for each account owner to review and sign.
6. Day three, agent, depth 2: flux analysis. Movements above threshold, explained from the transactions, sent to account owners to confirm.
7. Day four, agent, depth 1: close status and tie-outs. Checklist status, unsigned reconciliations, open items and a final subledger-to-ledger tie.
8. Day five, person, controller: reviews and signs. Then locks the period.

What does the controller review and sign?

Three things stay with the controller whatever the agent's depth. Judgements: provisions, estimates, write-offs, anything where two competent accountants could reach different numbers. Exceptions: reconciling items that are old or unexplained, intercompany differences no one has agreed, flux movements without a confirmed reason. And the close itself: the signature that says the books are fit to report.

Give the controller a single close pack from the agent: everything posted by the agent, everything parked and approved, everything still open, with the name of the person who approved each item. The month-end close checklist with agent tasks is a form you can adapt for this.

Which controls matter, and what evidence does each leave?

Preparer and reviewer are different. The agent prepares; a named person reviews. Evidence: identities on every reconciliation and journal.

Templates approved before close. No template changes during close week. Evidence: template versions with approval dates.

Cut-off enforced. The agent stops posting at the controller's cut-off. Evidence: timestamps on agent postings.

Close pack signed. Evidence: the signed pack with its item list.

Period lock by a person. Evidence: the lock event in the ledger, under the controller's identity.

What should you measure?

• Close days, before and after, but only alongside quality.
• Post-close adjustments. Rising adjustments mean speed has been bought with accuracy.
• Agent-prepared items rejected at review, by task.
• Reconciliations signed by day three.
• Audit findings on areas the agent prepares.

Where does it usually go wrong?

Review gets thinner. Drafts look finished, so reviewers sign faster. Keep a monthly sample where the reviewer re-performs one item in full.

Upstream lateness moves to the agent. Invoices still arrive late; the agent accrues more; reversals pile up. Fix the cut-off, not the accrual.

The calendar compresses before the controls catch up. A three-day close with the same number of reviewers means less review per item. Shorten the close only when review capacity allows.

Agent output treated as support. A reconciliation "supported" by an agent's schedule that was itself built from the ledger proves nothing. Support must come from outside the account being reconciled: a bank statement, a subledger, a third-party confirmation. The agent should state the source of its support on every reconciliation, and the reviewer should reject circular support.

Audit preparation is forgotten. Each month's evidence is what auditors ask for at year end. Keep it in a form that feeds the year-end audit request list.

Where Pryme fits

In Pryme, each close task is its own capability with its own depth, earned through Attest on your past closes. Posting recurring entries and parking judgemental ones are separate grants, and every verdict across close week is filed, so the controller's close pack can be rebuilt from the record. The finance bounds are in AI agents for finance.