How to automate bank reconciliation without losing the audit trail
How to automate bank reconciliation in six steps: clean inputs, written matching rules, classifying what's left, tie-out, review, and evidence for audit.
To automate bank reconciliation, connect bank and ledger data, check both inputs are complete, match lines with written rules in passes (exact first, then timing and references), classify what's left, prove the adjusted balances agree to zero, then route it for review. Keep the rule behind every match and the logic behind every total so it can be reproduced.
Dek: Six steps from exported statements to a reconciliation your reviewer signs in minutes and your auditor can reperform.
22 Oct 2026 · 9-minute read · Yoraito team
Why automate the bank rec first
Cash is the best place to start automating the close. Volume is high, the rules are clear and the result is easy to check: the reconciliation ties or it doesn't. Every other reconciliation depends on cash being right, and auditors test it every year.
The risk is automating it in a way nobody can explain. "The system matched it" is not an answer a reviewer or auditor can test.
Step 1: Connect the data (and stop pasting)
Pull bank activity through a bank feed or statement file, and GL cash detail from your ledger, for the same account and the same dates. Manual copy and paste is where rows get dropped.
Check before going further: same account, same start and end dates, and the bank ending balance taken from the statement itself.
Step 2: Prove the inputs are complete
Record, for each side:
| Check | Bank | Ledger |
|---|---|---|
| Rows | 48 | 52 |
| Total of activity | from the file | from the file |
| Opening balance + activity = closing balance? | Yes / No | Yes / No |
If either side doesn't roll forward from opening to closing, fix the data first. Matching incomplete data just produces confident wrong answers.
Step 3: Match with written rules, in passes
Run the strictest rule first, then loosen only dates and references, never amounts:
| Pass | Rule | Catches |
|---|---|---|
| 1 | Same amount, same date | Most lines |
| 2 | Same amount, dates within 3 days | Clearing delays |
| 3 | Same amount, same check number | Late-clearing checks |
| 4 | One bank line = sum of 2–5 ledger lines, same date | Batched deposits |
Each line can match once. Record which pass made each match: that's the answer to "why did these two match?" Deep dive on rules → /blog/ai-bank-reconciliation-matching-rules.
What about AI matching? AI can help suggest matches for messy descriptions. Treat its suggestions as proposals for a person, or turn them into new written rules once reviewed. Don't let a confidence score be the only reason two lines matched.
Step 4: Classify what's left
Every unmatched line belongs in one bucket:
- Bank only: fees, interest, unrecorded debits or deposits → record an entry.
- Ledger only: outstanding checks, deposits in transit → reconciling items that should clear next month.
- Errors: duplicates, transpositions → correct in the ledger.
Give each a date first seen and an owner → /glossary/close-and-reconciliation#reconciling-item.
Step 5: Prove the tie-out
Compute the adjusted bank balance (bank − outstanding checks + deposits in transit) and the adjusted book balance (ledger ± bank-only items ± corrections). The difference must be $0.00.
Worked example (Yoraito sample data):
| Amount | |
|---|---|
| Bank balance | $191,035.40 |
| Ledger balance | $180,932.80 |
| Difference before items | -$10,102.60 |
| Bank side: less outstanding check 4476 | -$3,640.00 |
| Book side: less wire fee not in ledger | -$25.00 |
| Book side: less ACH debit not in ledger | -$1,312.40 |
| Book side: add back check 4471 posted twice | +$7,800.00 |
| Adjusted bank balance | $187,395.40 |
| Adjusted book balance | $187,395.40 |
| Difference after items | $0.00 |
Never force a zero. If something is left, it's an open item with an owner.
Step 6: Review, sign and keep the evidence
The reviewer should see the rules, the items and the totals, not just a green tick. Keep, per account per month:
- Input profiles (rows and totals)
- Matches with the pass that made each
- Classified items with owners
- The tie-out calculation
- Preparer and reviewer sign-off with dates
With that set, an auditor can reperform the reconciliation independently. In Yoraito, every one of those figures is produced by a saved SQL query, so rerunning it gives the same result.
Three ways to automate, by team size
| Option | Good for | Evidence |
|---|---|---|
| Excel with formulas | 1–2 accounts, low volume | Formulas visible; manual matching for exceptions → /resources/bank-reconciliation-excel-template |
| Claude with a skill, on exports | Testing the method; small teams | Free Bank Rec Tie-Out skill writes a Trace sheet of every query → /claude-skills/bank-rec-tie-out |
| Reconciliation software | Many accounts, entities or clients | Yoraito: rules on every match, SQL behind every total, 22 named systems → /solutions/bank-reconciliation-software |
> Try the method free. The Bank Rec Tie-Out skill for Claude runs these six steps on your exports. [Get the skill →] /claude-skills/bank-rec-tie-out
Frequently asked questions
Can bank reconciliation be fully automated?
Matching, classification and the tie-out can be. Identifying unknown transactions and approving entries should stay with a person.
What's the best way to match bank transactions?
In passes, strictest first: exact amount and date, then small date windows and check numbers, then one-to-many. Never match on approximate amounts.
How often should we reconcile bank accounts once automated?
Weekly is practical once it's automated, which makes month-end mostly done before it starts.
Will my auditor accept an automated bank reconciliation?
If they can see the rules, the items and the calculation, and can reproduce the result, there's no reason not to. Questions auditors ask → /blog/ai-reconciliation-audit-questions.