Reconciliation automation is the use of software to match the transactions in one record against another, a bank statement against the cash ledger, a subledger against the general ledger, one entity's intercompany balance against the other's, and to leave a person only the items that don't match. The matching part is old and works well. Most accounting systems and bank feeds match transactions that agree on amount, date and reference, and cash application software, which does the same kind of matching on incoming payments with machine learning added to the rules, shows how far matching software goes: The Hackett Group found a median auto-match rate of 70% among end users, with a third exceeding 80%. Even that part is less widespread than it could be: Ventana Research found in 2023 that only 31 percent of organizations automate most or all of their reconciliations. And the time is real: APQC's benchmarks put the time to reconcile a single bank account at about 2.6 hours for top performers, 4 hours at the median and 5 or more for the bottom group, with manual matching and exceptions named as the causes. The exceptions are the part that rules leave behind, and they are where the newer work on reconciliation automation is aimed.
What account reconciliation covers in a monthly close
The US federal standard for internal control, the Government Accountability Office's Green Book, describes reconciliations as comparing transactions from two sets of records to determine whether they were recorded properly, have yet to be recorded, or were recorded improperly, and its example is the bank reconciliation. In practice a reconciliation confirms that the balance in a general ledger account agrees with an independent record of the same thing, and explains every difference between the two. A mid-market close carries five kinds. Bank reconciliations tie each cash account to the bank statement. Card reconciliations tie company card spending to the card statement and the receipts behind it. Subledger reconciliations tie the receivables, payables, inventory and fixed asset subledgers to their control accounts. Balance sheet reconciliations support every other material account, from prepaid expenses to accrued liabilities and clearing accounts. Intercompany reconciliations tie what one entity says it owes another to what the other says it is owed.
Each kind has the same structure: two records, a set of items that should correspond, and a residue that doesn't. What differs is where the independent record comes from and how the items are identified. A bank statement identifies a deposit by amount, date and whatever reference the payer typed. A subledger identifies an invoice by number. The quality of those identifiers decides how much of the reconciliation a rule can do.
The matching rules that clear most of the volume
Matching rules come in four shapes. One to one: a single bank line matches a single ledger entry on amount and date, with a reference to confirm it. One to many: a single deposit settles several invoices, and the rule has to find the combination of open items that sums to the deposit. Many to one: several partial payments settle a single invoice. And tolerance matches: two items differ by a small amount, such as a bank fee or a rounding difference, and the rule accepts the match and books the difference.
The volume is large enough to make the shapes matter. The Federal Reserve counted 236.6 billion noncash payments in the United States in 2024, including 9.2 billion checks, and each one lands in someone's ledger and someone's bank statement. The first shape is easy and fast. The second is where rules begin to struggle, because the number of possible combinations grows quickly with the number of open items, and a rule that tries them all will find coincidental matches that are wrong. Good reconciliation software narrows the search with references, customer identity and dates. Where the references are missing, the rule can't tell which combination is right, and the item goes to a person.
What sits in the unmatched remainder
The remainder has a small number of causes, and they recur every month. Timing differences are the most familiar: a deposit recorded on the last day that the bank credits on the first day of the next month, a check written and recorded but not yet cashed. Combined and net settlements come next: a card processor pays one amount for a day's sales net of fees and chargebacks, and a customer pays twenty invoices with one transfer and no remittance. Missing or wrong references produce a third group, where the money is right and nothing identifies what it is for. The rest are genuine errors: a duplicate entry, a transposed amount, an invoice posted to the wrong customer, a payment recorded twice.
The first two groups are explainable from documents the finance team already has, such as the processor's settlement report, the customer's remittance email or the next month's bank statement. The accountant spends the hours finding and reading those documents. The errors are the part that needs an accountant's judgment, because correcting them means deciding which record is wrong and what entry fixes it.
Why the remainder is what an auditor reads
A reconciliation is a control, and the remainder is the part of it that carries the risk. An unexplained difference carried forward from month to month means the ledger and the independent record disagree and nobody knows why, which is the condition in which misstatements, misappropriations and simple errors persist. Auditors test reconciliations for exactly this reason, and the PCAOB's standards list reconciliations and the follow-up of reconciling items among the controls an auditor considers. The failures show up in public filings. KPMG's study of fiscal 2025 annual reports found that 238 of 3,307 companies disclosed material weaknesses in internal control, and roughly one in ten of those cited untimely or inadequate account reconciliations, with the financial close and reporting process involved in 84 percent of them. They look at whether the reconciliation was done on time, whether the reconciling items were explained with support, whether old items were cleared, and whether a reviewer signed off. A reconciliation that clears most lines by rule and carries a long tail of stale items is a weak control, however high its match rate.
What reconciliation automation looks like with an agent in the loop
Gartner has described the automated close in the same terms, with software performing the reconciliations and sending the exceptions and errors to a person. An agent working a reconciliation starts where the rules stop. For each unmatched item, it gathers the documents that explain it: the processor's settlement report for a net deposit, the remittance email or portal record for a combined customer payment, the next statement for a timing difference, the invoice and the posting history for a suspected error. It proposes a match or an adjusting entry and attaches the evidence, so the accountant reviewing the proposal can check it in a minute against the source. Items it can't explain go to the accountant as open questions, with what the agent looked at and ruled out.
The same approach applies across the five kinds of reconciliation, with different sources. For a card reconciliation the evidence is the receipt and the expense policy; for intercompany it is the other entity's ledger and the agreement behind the charge. The accountant keeps the decisions that change the books, which are the correcting entries, the write-offs and the sign-off. The practical effect is on the close, since reconciliations sit on its critical path. We describe the sequence in the month-end close, rebuilt, and the receivables side of the same matching problem in rebuilding cash application.
A reasonable counter, answered
A reasonable counter is that the accounting system already does this: QuickBooks, NetSuite and the other mid-market systems match bank feed transactions automatically, and the finance team clicks through the suggestions. For the one-to-one matches, they do, and a team should use them. The counter mistakes the matched lines for the job. The hours in a reconciliation go into the remainder, the combined payments, the net settlements, the missing references and the errors, and a bank feed's suggestion engine leaves all of those to the accountant, usually without the documents that would explain them. That remainder sets the cost of the process, as it does in any automation, which is the argument we make in the unit economics of the exception tail. Automating the easy lines was the first half of reconciliation automation. The second half is the investigation, and that is the half a finance team still does by hand.
Common questions
- What is account reconciliation?
- Account reconciliation is the control that confirms the balance in a general ledger account agrees with an independent record of the same thing, such as a bank statement, a subledger, a card statement or the other party's books. Every difference between the two has to be explained and, where it is an error, corrected. Reconciliations are done at least monthly for material accounts, and they are what an auditor checks first when a balance looks wrong. A reconciliation with an unexplained difference carried forward month after month is a sign the control isn't working.
- How do you do an account reconciliation?
- Start with the ledger balance and the independent balance at the same date, then match the individual items on each side: deposits to receipts, payments to disbursements, subledger entries to their postings. Items that appear on one side only are then sorted into timing differences, such as deposits in transit and outstanding checks, and genuine differences, such as errors, duplicates and missing entries. Timing differences are listed and checked in the following period, and genuine differences are corrected with a journal entry. The reconciliation is finished when the adjusted balances agree and a reviewer has signed off the explanation.
- What can be automated in a bank reconciliation?
- The matching of transactions that agree on amount, date and reference can be automated almost entirely, and most accounting systems and bank feeds already do it. The work that remains is the unmatched remainder: a deposit that combines many customer payments, a card processor settlement net of fees, a payment with no reference, a timing difference that crosses the month end. An agent can investigate that remainder by reading remittances, statements and supporting documents, and propose a match or an adjusting entry with the evidence attached. The accountant reviews and approves the proposals, and keeps the judgment on errors and write-offs.