Bank Reconciliation: What It Is, How to Automate It, and Why AI Changes Everything

Bank reconciliation is the process of comparing a company's cash records in the general ledger against the bank statement to verify that both reflect the same transactions and the same ending balance. It is the most fundamental form of account reconciliation and the one that every accounting team performs, every month, for every bank account.

The concept is simple. The execution, at scale, is not. A mid-market company with 15 bank accounts processes tens of thousands of transactions per month across those accounts. Each transaction in the bank statement should have a corresponding entry in the GL. Each entry in the GL should have a corresponding transaction in the bank statement. When they do not match, someone has to figure out why.

This guide covers the complete bank reconciliation process: what it is, why it matters, how it works step by step, what the common reconciling items are, and how AI is transforming the process from a manual matching exercise into an autonomous workflow.

What is bank reconciliation?

Bank reconciliation is the comparison of two records of the same cash activity: the company's general ledger (what the company recorded) and the bank statement (what the bank recorded). The two records should agree. When they do not, the differences are reconciling items that must be identified, investigated, and documented.

The general ledger reflects what the company knows about: checks written, deposits made, electronic payments sent, wire transfers received. The bank statement reflects what the bank has processed: checks that have cleared, deposits that have settled, fees that have been assessed, interest that has been credited.

The two records differ for legitimate reasons. A check written on March 29 may not clear the bank until April 3. A deposit made on March 31 may not appear on the bank statement until April 1. These timing differences are normal and expected. The reconciliation identifies them, documents them, and confirms that the underlying cash balance is correct.

Why it matters

Bank reconciliation matters for three reasons.

Accuracy. Cash is the most liquid and most susceptible-to-error asset on the balance sheet. A bank reconciliation catches posting errors (wrong amount, wrong account, duplicate entries), timing differences, and unauthorized transactions before they flow into the financial statements.

Fraud detection. Unauthorized payments, check tampering, and electronic fraud surface during bank reconciliation. When a payment appears on the bank statement that nobody authorized, the reconciliation process identifies it. The longer reconciliation is delayed, the longer fraud goes undetected.

Control environment. Auditors and regulators consider bank reconciliation a key internal control. SOX-regulated companies must demonstrate that bank reconciliations are performed timely, reviewed by an independent party, and that reconciling items are resolved within a reasonable period. An overdue bank reconciliation is an audit finding.

The bank reconciliation process: step by step

The process follows a consistent sequence regardless of account volume or transaction complexity.

Step 1: Gather data

Obtain the bank statement for the reconciliation period (typically month-end). Obtain the GL cash ledger detail for the same period. Both should show the same date range and the same ending date.

For automated reconciliation, this step is handled by bank feed integration. The bank transmits transaction data electronically (via BAI2 file, MT940, or API) directly into the reconciliation platform. No manual download required.

Step 2: Compare ending balances

Note the GL ending balance and the bank statement ending balance. The difference between them is the total reconciling amount that must be explained.

Example:

Record	Ending balance
General ledger	$4,287,500.00
Bank statement	$4,312,750.00
Difference	$25,250.00

The $25,250 difference does not mean someone made a mistake. It means there are transactions recorded in one system that have not yet been recorded in the other. The next steps identify what those transactions are.

Step 3: Identify bank-side reconciling items

Review the bank statement for items that appear in the bank's records but not in the GL. These are transactions the bank has processed that the company has not yet recorded.

Common bank-side items:

• Bank fees. Monthly service charges, wire transfer fees, lockbox fees, returned check fees. The bank deducts these from the account, but the company may not record them until the bank statement is received.
• Interest earned. Interest credited by the bank that the company has not yet recorded as income.
• Direct deposits received. Customer payments made directly to the bank (ACH, wire) that the company has not yet posted.
• NSF (returned) checks. Checks deposited by the company that bounced. The bank reverses the deposit, but the company's GL still shows the original deposit.
• Bank errors. Rare, but they happen. A transaction posted to the wrong account, a duplicate fee, an incorrect amount.

Each of these items requires a journal entry to update the GL. Bank fees are recorded as an expense. Interest is recorded as income. NSF checks reverse the original deposit and reinstate the receivable.

Step 4: Identify book-side reconciling items

Review the GL for items that appear in the company's records but not on the bank statement. These are transactions the company has recorded that the bank has not yet processed.

Common book-side items:

• Outstanding checks. Checks written and recorded in the GL but not yet presented to the bank for payment. This is typically the largest category of reconciling items for companies that still issue checks.
• Deposits in transit. Deposits made near period end that the company recorded but the bank had not yet credited. With same-day ACH and electronic deposits, this category is shrinking but still relevant for physical deposits.
• Electronic payments in transit. ACH payments or wire transfers initiated near period end that have not yet settled.
• Recording errors. Transactions posted to the GL with the wrong amount, date, or account. A check for $1,245 recorded as $1,254 creates a $9 reconciling item.

Outstanding checks and deposits in transit do not require journal entries. They are timing differences that will clear when the bank processes them. Recording errors do require correcting entries.

Step 5: Record adjustments

Post journal entries for all items that require GL updates. This typically includes:

Adjustment	Debit	Credit
Bank service fees	Bank charges expense	Cash
Interest earned	Cash	Interest income
NSF check	Accounts receivable	Cash
Recording error correction	(Depends on error)	(Depends on error)

After posting these entries, the adjusted GL balance should equal the adjusted bank balance. If it does not, there are unidentified reconciling items that require further investigation.

Step 6: Document and certify

Produce the reconciliation work paper. The standard format is:

Bank to book reconciliation:

Item	Amount
Bank statement ending balance	$4,312,750.00
Less: outstanding checks	($38,250.00)
Plus: deposits in transit	$15,000.00
Less: bank error (duplicate fee)	($2,000.00)
Adjusted bank balance	$4,287,500.00

Item	Amount
GL ending balance per books	$4,287,500.00
Less: bank fees not yet recorded	($1,200.00)
Plus: interest earned	$450.00
Less: NSF check	($3,500.00)
Plus: recording error correction	$4,250.00
Adjusted book balance	$4,287,500.00

Both adjusted balances agree. The reconciliation is complete. The preparer signs off, the reviewer validates, and the work paper is filed.

Common reconciling items explained

Understanding the common reconciling items accelerates the investigation process. Here is a detailed breakdown of what causes the most frequent differences.

Outstanding checks

Checks written by the company and recorded in the GL but not yet cashed or cleared by the bank. These appear in the GL but not on the bank statement.

Outstanding checks are normal in the first few days after period end. Checks outstanding for more than 30 days may indicate a lost check, a vendor that has not deposited it, or a check that was voided in the system but the void was not recorded. Checks outstanding for more than 90 days may need to be escheated (turned over to the state as unclaimed property) depending on jurisdiction.

Best practice: Maintain an outstanding check register. Age the items. Investigate any check outstanding more than 60 days. Void and reissue if necessary.

Deposits in transit

Cash or checks deposited at the bank near period end that have not yet been credited to the account. The company's GL shows the deposit, but the bank has not processed it.

With electronic banking, deposits in transit are less common than they used to be. ACH transfers and wire deposits clear same-day or next-day. Physical check deposits may take 1 to 3 business days to clear.

Bank errors

Banks make mistakes. A deposit credited to the wrong account. A fee assessed twice. A wire transfer posted with the wrong amount. These are rare but consequential.

When you identify a bank error, contact the bank immediately. The bank will issue a correction, but it may take 1 to 3 business days. The reconciling item should be documented with the bank reference number and expected resolution date.

Book errors

More common than bank errors. A transaction posted to the wrong account (coded to operating cash instead of payroll cash). A payment recorded with a transposed amount ($1,245 vs. $1,254). A duplicate entry from a system glitch.

Book errors require correcting journal entries. Document the error, the correction, and the root cause. If the same error recurs, it indicates a systemic issue (bad interface, manual keying, missing validation) that should be fixed at the source.

Bank fees and charges

Monthly service fees, wire transfer fees, overdraft charges, lockbox fees, foreign transaction fees. The bank deducts these from the account and reports them on the statement. The company may not know the exact amounts until the statement arrives.

Best practice: Estimate bank fees monthly and record an accrual. True up when the actual statement arrives. This prevents the fees from appearing as a surprise reconciling item.

Interest earned

Interest credited by the bank on interest-bearing accounts. Similar to fees, the exact amount may not be known until the statement arrives. Record as income when identified.

Foreign currency items

For bank accounts denominated in foreign currencies, the GL balance (translated to reporting currency) will differ from the bank statement balance (in local currency) due to exchange rate movements. These differences are not reconciling items in the traditional sense. They are currency translation adjustments that should be recorded as part of the period-end revaluation process.

Manual vs. automated bank reconciliation

The time difference between manual and automated bank reconciliation is dramatic.

Dimension	Manual	Rules-based auto	AI-native auto
Data gathering	30-60 min (download, format, import)	Automatic (bank feed)	Automatic (bank feed)
Transaction matching	2-4 hours (line by line)	10-20 min (rules run, exceptions flagged)	5-10 min (AI matches and investigates)
Exception investigation	1-3 hours	1-3 hours (same manual effort)	15-30 min (AI investigates, human reviews)
Work paper creation	30-60 min	15-30 min (template-based)	Automatic
Total per account	4-8 hours	1.5-4 hours	15-30 min (review only)
Auto-match rate	0%	40-60%	85%+
Scalability	Linear (more accounts = more FTEs)	Semi-linear	Near-flat (AI scales with volume)

For a company with 20 bank accounts, the math is simple:

• Manual: 80 to 160 hours per month. 2 to 4 FTEs dedicated to bank reconciliation.
• Rules-based: 30 to 80 hours per month. 1 to 2 FTEs.
• AI-native: 5 to 10 hours per month. Part of one person's responsibilities.

Transaction matching approaches

The matching engine is the core of automated bank reconciliation. Three approaches exist, each with different capabilities and limitations.

One-to-one matching

The simplest form. One bank transaction matches one GL entry. Amount, date, and reference number all agree. This handles the straightforward cases: a check cleared, a wire received, a recurring payment processed.

Rules-based systems handle one-to-one matching well. The limitation is that many real-world transactions do not fit this pattern.

One-to-many and many-to-one matching

A single bank transaction matches multiple GL entries, or multiple bank transactions match a single GL entry.

Example: A company deposits three customer checks ($5,000, $3,200, and $1,800) as a single bank deposit of $10,000. The bank shows one transaction. The GL shows three. The matching engine must recognize that 5,000 + 3,200 + 1,800 = 10,000 and create a compound match.

Rules-based systems can handle simple one-to-many matches using amount summation within a date range. They struggle when the amounts do not sum exactly (rounding, fees, partial payments).

Many-to-many matching

Multiple bank transactions match multiple GL entries with no clean one-to-one or one-to-many relationship.

Example: Five customer wire payments totaling $87,450 on the bank statement correspond to eight AR invoice payments totaling $87,437.50 in the GL, with the $12.50 difference being a wire transfer fee netted by the bank. Rules-based systems cannot solve this. They flag the entire cluster as unmatched.

AI-native platforms handle many-to-many matching by analyzing transaction attributes (dates, amounts, descriptions, counterparties) and finding the optimal grouping that explains the differences. The $12.50 fee is identified, documented, and included in the match explanation.

How AI transforms bank reconciliation

AI changes bank reconciliation in three fundamental ways.

1. Intelligent matching beyond rules

Traditional matching engines use deterministic rules: match if amount equals, match if amount is within tolerance, match if reference number equals. AI uses contextual reasoning. It considers amount, date, description, counterparty, historical patterns, transaction frequency, and related transactions to determine matches.

A monthly rent payment of $12,500 that appears on the bank statement as "WIRE TRF - PROPERTY MGMT" and in the GL as "Rent Expense - Office Lease" is an easy match for a human but a difficult match for a rules-based engine (different descriptions, potentially different dates). AI recognizes the pattern after seeing it once.

2. Autonomous exception investigation

When a transaction does not match, rules-based systems flag it and wait for a human. AI investigates.

For an unmatched bank fee of $47.50, AI checks: Is this a known recurring fee? Does the amount match prior months? Is there a GL account where bank fees are typically recorded? Has a similar fee been reconciled before? Based on the investigation, AI either identifies the match in a different account, proposes a journal entry to record the fee, or escalates with a documented explanation of what it found and what it could not resolve.

This is the capability described in our AI-powered bank reconciliation deep dive.

3. Work paper generation with evidence

Manual work papers are inconsistent. Some preparers document thoroughly. Others write "timing difference" and move on. AI-generated work papers are comprehensive by default. Every match includes the matching criteria. Every reconciling item includes the investigation findings. Every exception includes the confidence level and recommended action.

The result is audit-ready documentation produced as a byproduct of the reconciliation process, not as a separate manual step.

How Arvexi handles bank reconciliation

Arvexi's account reconciliation platform handles bank reconciliation as an AI-native workflow from data ingestion to certified work paper.

Bank feed integration. Bank transaction data flows into Arvexi automatically via BAI2, MT940, or direct API connection. GL cash ledger data imports from your ERP. Both datasets are mapped to the reconciliation template for each bank account.

AI-powered transaction matching. Cortex AI applies multi-criteria matching across the bank and GL datasets. One-to-one, one-to-many, many-to-one, and many-to-many matches are all handled. The matching considers amount, date, description, counterparty, and historical patterns. Match rates consistently exceed 85% on well-integrated accounts.

Exception investigation. Unmatched items are not simply flagged. Cortex investigates each exception: checking for timing differences, identifying fees and charges, looking for amount discrepancies that might indicate a recording error, and cross-referencing other accounts for mispostings. Each investigation produces a documented finding.

Automatic work paper generation. For every bank account, Cortex produces a complete reconciliation work paper: GL balance, bank balance, all matched transactions, all reconciling items with explanations, proposed adjusting entries (if applicable), and the net reconciled difference. The work paper meets the documentation standard auditors require.

Confidence-based certification. Accounts reconciled with high confidence are auto-certified. Accounts with exceptions are routed for review with the full investigation context. Your team spends their time on the items that genuinely require human judgment, not on the mechanical matching that AI handles better and faster.

Common bank reconciliation mistakes

These are the mistakes that cause audit findings and operational problems.

Reconciling old months instead of staying current. When reconciliation falls behind, the team is always working on a prior month while the current month's problems compound. Break the cycle by automating the current month first, then clearing the backlog.

Treating outstanding checks as permanent. Outstanding checks should clear. If a check has been outstanding for 90+ days, investigate. It may need to be voided and reissued, or escheated to the state as unclaimed property.

Not segregating bank account purposes. Reconciling a single operating account that handles payroll, vendor payments, customer receipts, and tax payments is harder than reconciling four purpose-specific accounts. The transaction mix is cleaner when accounts are segregated by purpose.

Ignoring small differences. A $3 difference might be immaterial. But if $3 differences appear every month, the aggregate becomes material and the root cause (likely a rounding issue or fee) should be identified and resolved.

Reconciling to the wrong statement. Using the previous month's bank statement. Using an interim statement instead of the month-end statement. Using a statement from the wrong account. These errors invalidate the entire reconciliation and waste hours of investigation time. Automated bank feeds eliminate this class of mistake.

Not reviewing the outstanding check list. The outstanding check list grows over time if nobody reviews it. Stale checks accumulate. The list becomes unreliable. Review the outstanding check list monthly, investigate items older than 60 days, and void/reissue as necessary.

Bank reconciliation for specific account types

Operating accounts

The highest-volume bank accounts. Hundreds or thousands of transactions per month including vendor payments, customer receipts, payroll, tax payments, and intercompany transfers. These accounts benefit most from AI-powered matching because of the volume and variety of transactions.

Payroll accounts

Dedicated bank accounts funded before each payroll run. The funding transfer and individual direct deposit debits should match the payroll register. Reconciliation is relatively straightforward because the payroll system provides a detailed register of every payment. Common exceptions: returned direct deposits, manual checks for employees not on direct deposit, and payroll tax debits.

Accounts receivable lockbox accounts

Collections accounts where customer payments are deposited. Reconciliation matches bank deposits against the AR sub-ledger. The challenge: customer remittance information may not clearly identify which invoices are being paid, especially for customers who batch multiple invoices into a single payment.

Foreign currency accounts

Bank accounts denominated in currencies other than the reporting currency. The bank statement balance (in local currency) is reconciled normally. Then the reconciled balance is translated to reporting currency at the period-end rate for the balance sheet. Translation gains/losses are recorded as part of the period-end revaluation.

Petty cash

Low-volume, low-value accounts that are still reconciled for control purposes. The GL balance is reconciled against a physical cash count and supporting receipts. Petty cash reconciliation is often manual because the transaction volume does not justify automation.

Getting started with automated bank reconciliation

If your team reconciles bank accounts manually in spreadsheets, here is the implementation path.

Week 1: Connect bank feeds. Establish automated bank statement delivery (BAI2, MT940, or API) for your highest-volume accounts. Connect your ERP GL data feed.

Week 2: Configure matching. Set up matching rules and train the AI on your transaction patterns. Run the first automated reconciliation alongside your manual process and compare results.

Week 3: Parallel run. Run both processes for one full close cycle. Verify that automated matches are correct. Review AI-investigated exceptions. Confirm that work papers meet your documentation standards.

Week 4: Go live. Cut over to AI-powered reconciliation. Your team reviews automated work instead of building it manually. Monitor match rates and exception volumes. Fine-tune matching configuration based on the first production cycle.

Explore Arvexi's bank reconciliation or request a demo to see AI-powered transaction matching with your industry's data.

For related reading, see our guides on balance sheet reconciliation, account reconciliation software, and the complete R2R guide.