Payment reconciliation exceptions
PSP settlement files matched to the ledger; small breaks fix themselves, large ones go to finance approval.
Payment reconciliation exceptions is an agent scenario in the Fintech & Payments sector running on the batch, api, email channels. Trigger: Nightly batch: PSP/acquirer T+1 settlement file received. The Reconciliation Supervisor coordinates 3 specialist agents (Matching Agent, Investigation Agent, Correction Agent) across 5 systems (PSP / Acquirer API, Card Network, Core Ledger, Case Management, Email); of 10 steps, 2 write to systems and 2 are checks or approvals. Outcome: All exceptions matched or corrected, journals posted, open items with finance. Approval rule: Single break > 10,000 or total > 50,000 → finance manager approval before adjustment.
Payment reconciliation exceptions
- Settlement file received: 184,320 transactions, 41,870,500, T+1
- PSP detail pulled; 99.86% matched, 258 breaks: 143 fees, 61 refunds, 39 FX, 15 missing
- Plan: root-cause classification → tolerance check → auto-fix → finance approval
- Card network fee schedule pulled: 143 lines explained by a scheme fee change
- Ledger entries retrieved: 61 refunds booked on T+2 → timing difference
- Tolerance: 243 exceptions ≤ 500 → auto-fix; 15 missing transactions 186,400 → finance
- 243 correcting journals posted (fees and timing), total 21,340
- Break > 10,000 → finance manager approved the adjustment for 15 missing transactions
- Case #REC-1188 opened, missing-transaction query sent to the PSP
- Summary email to finance: 243 auto-corrected, 15 items in an open case
- ■completed
Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus
Nightly batch: PSP/acquirer T+1 settlement file received
All exceptions matched or corrected, journals posted, open items with finance
Agents
Reconciliation Supervisor
Decomposes the objective, delegates to agents, manages approval points, merges the result.
Matching Agent
Matches settlement lines to ledger entries, isolates exceptions
parse_settlement_filematch_transactionslist_exceptionsInvestigation Agent
Determines the root cause: fees, FX, refunds, timing
get_psp_transactionget_ledger_entryget_card_network_feesclassify_breakCorrection Agent
Posts in-tolerance correcting journals, routes the rest to finance
post_journalcreate_exception_casesend_emailSteps
| # | Kind | Agent | What happens | System | ms | tok |
|---|---|---|---|---|---|---|
| 01 | ingest | Reconciliation Supervisor | Settlement file received: 184,320 transactions, 41,870,500, T+1 | — | 300 | — |
| 02 | tool | Matching Agent | PSP detail pulled; 99.86% matched, 258 breaks: 143 fees, 61 refunds, 39 FX, 15 missing | PSP / Acquirer API | 1,400 | — |
| 03 | plan | Reconciliation Supervisor | Plan: root-cause classification → tolerance check → auto-fix → finance approval | — | 280 | 200 |
| 04 | tool | Investigation Agent | Card network fee schedule pulled: 143 lines explained by a scheme fee change | Card Network | 540 | — |
| 05 | retrieve | Investigation Agent | Ledger entries retrieved: 61 refunds booked on T+2 → timing difference | Core Ledger | 480 | — |
| 06 | verify | Investigation Agent | Tolerance: 243 exceptions ≤ 500 → auto-fix; 15 missing transactions 186,400 → finance | — | 760 | 820 |
| 07 | write | Correction Agent | 243 correcting journals posted (fees and timing), total 21,340 | Core Ledger | 680 | — |
| 08 | approval | Reconciliation Supervisor | Break > 10,000 → finance manager approved the adjustment for 15 missing transactions | — | 3,200 | — |
| 09 | write | Correction Agent | Case #REC-1188 opened, missing-transaction query sent to the PSP | Case Management | 440 | — |
| 10 | notify | Correction Agent | Summary email to finance: 243 auto-corrected, 15 items in an open case | 260 | — |
Merchant operations assistant: POS, settlement, refund and limit
The merchant writes on WhatsApp; POS fault, settlement query, refund and limit increase are resolved in one dialogue.
Merchant KYB onboarding
Registry, UBO and sanctions checks in one flow; low-risk merchants start accepting payments the same day.
AML transaction monitoring alert triage
Alerts arrive enriched with a draft narrative; the analyst always decides and the filing goes out in minutes.
Time to move from experimenting with AI to transforming with it.
In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer.
- Your 2–3 priority problems
- Live demo of a similar scenario
- Roadmap starting with value analysis