Reconciliation Optimization Agent
Match payment and financial evidence, classify exceptions, and focus operators on unresolved monetary impact.
On this page
Automate reconciliation across your payment stack.
Bring order, processor, acquirer, network, settlement, payout, bank, fee, refund, and dispute evidence into one operating view—then explain the exceptions that still need attention.
- Evidence matching
- Exception classification
- Resolution priority
What the agent does#
- Normalize identifiers, timestamps, currencies, signs, amounts, and record grain across sources.
- Match records using configured keys, tolerances, lineage, and settlement relationships.
- Classify missing, delayed, duplicated, mismatched, or fee-related exceptions.
- Rank exceptions by age, monetary impact, confidence, provider, and operational owner.
- Produce the evidence and recommended next step required for investigation or adjustment.
- Track the resolution and use the confirmed outcome to improve later classification.
Evidence model#
| Layer | Representative evidence |
|---|---|
| Commerce | Order, invoice, merchant reference, amount, currency |
| Payment | Authorization, capture, refund, processor response, gateway transaction |
| Settlement | Batch, ARN, gross and net amount, interchange, scheme and provider fees |
| Bank | Deposit, value date, bank reference, payout, remittance information |
| Operations | Dispute, reserve, exception owner, resolution, accounting adjustment |
See Payment Reconciliation for the field-level data map.
Automation boundary#
Matching and classification do not by themselves authorize an accounting entry, provider adjustment, refund, dispute response, or write-off. Define which outcomes are informational, which can update an operational status, and which require finance or operations approval.
When the same evidence supports more than one plausible match, retain the candidates and confidence instead of forcing a reconciliation. A manually resolved exception must preserve the operator, reason, evidence, and time of the decision.
Measure the operation#
- automatic match rate and amount;
- exceptions by type, provider, age, and monetary impact;
- time to investigate and time to resolve;
- recovered or corrected value;
- false-match and reopened-exception rate;
- source freshness, missing files, and ingestion failures.