Can an AI agent prepare balance sheet reconciliations?

Yes, for every account, every month. An AI agent can prepare balance sheet reconciliations by substantiating each balance against its supporting detail, ageing reconciling items and drafting corrections. The account's risk decides the review. Low-risk accounts can be sampled; cash, suspense, tax and intercompany need a named reviewer who signs each one. The agent never signs its own work.

What does a balance sheet reconciliation prove?

That the number on the balance sheet is supported by something outside the ledger account itself. A prepayments balance is supported by the prepayment schedule. An accrued expenses balance by the list of accruals and their evidence. A VAT control account by the return and payments. If the supporting detail does not add up to the ledger balance, something is wrong, and the difference is what the reconciliation explains.

Many teams reconcile their important accounts monthly and leave the rest for quarter or year end. That is where errors build up. An agent removes the capacity problem: it can prepare every account every month, so the question becomes how much review each needs.

What conditions decide the review level?

Each account gets a risk rating, set by the financial controller, based on:

• Size and volatility. Large balances and accounts that move a lot.
• Judgement in the balance. Provisions and estimates are higher risk than a schedule-driven balance.
• History. Accounts that have had errors before.
• Sensitivity. Cash, suspense, intercompany, tax and anything a person could use to hide a transfer.

The rating sets the review. The detailed method is in how to prepare balance sheet reconciliations with an AI agent.

What does the agent prepare, and where does a reviewer sign off?

Separation of duties comes from the reviewer column. The reviewer must not be the person who posts to the account, and the agent's corrections are drafted for someone with posting rights to approve. A reconciliation reviewed by the person who controls the account proves little.

An illustrative example

Take a retailer with 220 balance sheet accounts. At month end the agent reconciles all of them within two days of close.

For 190 low-risk accounts, the supporting detail agrees and there are no items older than 60 days. The management accountant reviews a sample of 20 and signs them off. For 24 medium-risk accounts, three show differences: one gift card liability account is 3,400 lower than the gift card system says is outstanding. The agent traces it to a batch of redemptions recorded twice, drafts the correcting journal and routes it to the accountant.

Of the six high-risk accounts, the suspense account holds 41 items, nine of them older than 90 days. The agent ages each, attaches what it found about each item and lists the oldest first. The financial controller reviews them, assigns owners for clearing, and signs the reconciliation with those items open and explained.

What controls make it safe?

• Risk ratings owned by the controller. The agent cannot downgrade an account.
• No balancing entries. Differences are explained or reported, never plugged.
• Reviewer independence. Recorded on each reconciliation and checked against posting rights.
• Ageing thresholds. Items past the limit are escalated automatically.
• A standard format. Each reconciliation follows the same structure, such as the balance sheet reconciliation template, so reviewers know where to look.

How does this relate to other reconciliations?

Bank and intercompany reconciliations are balance sheet reconciliations with their own particular sources. Cash is matched to the bank, as set out in whether an AI agent can do bank reconciliations. Intercompany is matched across two ledgers, as covered in whether an AI agent can do intercompany reconciliations. The same rule runs through all three: the agent explains, a person signs. The wider controls sit in AI agents for finance: books you can audit.