Bank to ledger matching, with every break explained.
The Reconciliation Agent ingests bank and GL activity, scores candidate matches by amount, date, and reference, and classifies what is left: transpositions, timing differences, duplicates, and true exceptions. It proposes adjusting journals with cited lines — your team approves before anything posts.
How it works
The same pipeline you see in the live module, step by step.
- 1
Ingest bank and ledger
Parse BAI2, MT940, CAMT.053, or CSV bank files alongside GL cash detail. Normalize signs, dates, and references to a common matching key.
- 2
Auto-match within policy
Pair lines by exact amount, by reference token, and by fuzzy description within date windows you configure. One-to-many and many-to-one sets are scored and ranked.
- 3
Classify exceptions
Flag transpositions when differences divide by nine, deposits in transit when bank date follows ledger date, outstanding checks when only GL shows the payment, and duplicate payments when the same reference hits twice.
- 4
Propose journals
Draft adjusting entries with memo text that quotes bank and ledger amounts. Route to approvers by threshold and account sensitivity.
- 5
Close the period
Publish a reconciliation summary with matched balance, open items, and approved adjustments — ready for controller sign-off.
Worked example on demo data
Illustrative run on fictional Aster Bank data — not a customer result.
WIRE-88213 transposition on Aster Bank demo data
- September operating account activity includes outgoing wire WIRE-88213 to Halvorsen Trade Partners GmbH.
- The bank statement shows $15,382.00 debited on 18 Sep 2026. The GL cash line for the same reference was entered as $15,832.00 — a classic transposition of the tens and hundreds digits.
- The agent proposes a $450 adjusting entry once a treasury manager approves; until then the item stays in the exception queue with both amounts quoted in the trace.
| Source | Reference | Amount | Date |
|---|---|---|---|
| Bank (BAI2) | WIRE-88213 | $15,382.00 | 2026-09-18 |
| GL 1010 | WIRE-88213 | $15,832.00 | 2026-09-18 |
| Variance | Transposition | $450.00 | — |
Checks and rules
Deterministic logic on every action; optional intelligence enriches text when configured.
- Exact amount match within penny tolerance
- Reference token match (wire ID, lockbox, check number)
- Transposition heuristic when |bank − ledger| is divisible by 9
- Deposit in transit when ledger precedes bank within N days
- Duplicate payment detection on reference + amount
- Corporate card aggregation vs individual bank debits
What you get
Artifacts land in the shared finance record and activity log.
- Matched pairs with match type and confidence
- Exception queue with suggested disposition
- Proposed journal entries with approval routing
- Reconciliation summary for the account and period
- Activity log entries for each agent run
Integrations (built during pilot)
- BAI2 / MT940 / CAMT.053 bank files
- GL cash detail from SAP, Oracle, NetSuite, Workday, Dynamics
- Core banking transaction exports (Temenos, FIS, Fiserv, Finastra, Jack Henry)
- SFTP scheduling with file checksum validation
Where teams use Reconciliation Agent
All industriesQuestions
Does FinanceAI need live API access on day one?
No. Many pilots start with scheduled BAI2/MT940/CAMT.053 files and GL exports over SFTP. Read-only APIs come next once security review passes. The live demo uses the shared Aster Bank synthetic ledger.
What if Stellitron Intelligence is unavailable?
Every decision still runs through deterministic rules: amount matching, transposition checks, UBO percentage math, screening score thresholds, and flux materiality gates. Optional intelligence only improves extraction and narrative.
Who approves agent proposals?
Your org chart. Reconciliation journals route to treasury or accounting managers by amount. KYC dispositions require onboarding or BSA approval per your policy. Close entries respect maker-checker and each user's approvalLimit in the app.
Can it reconcile multiple bank accounts and entities?
Yes. Each account/entity combination runs its own match pass and summary. Intercompany pairs can be handed to the Close Agent when both legs sit in different ledgers.
How do you handle outstanding checks and DIT?
Timing rules compare bank value dates to GL posting dates and classify known patterns — for example lockbox LBX-48250 credited on 1 Oct while GL recorded 30 Sep. Users can accept the agent's timing classification or reclassify with a reason.