Payment investigations worked as one cited thread, from the request to the signed reply
Wire inquiries answered in hours, not days, with every liability reply checked.
See one case, screen by screen ↓Payment Investigations is software in which payment operations teams work wire inquiries, cancellations and non-receipt claims as cases, with replies. Each case reads as one thread: the request, the agents' research steps with the record each one read, and the drafted replies. The analyst sends routine replies; a team lead approves any reply that admits liability, moves money or pays a claim.
AI use cases in banking: a buyer's guide to agentic AI for bank and credit union operations →
Updated
Payment investigations are the questions that come back after a wire is sent or received: our customer was not credited, please cancel, where is my wire, we claim interest for the late credit. Before anyone can answer, the payment investigation analyst finds the payment in four to six systems — the payments hub, the SWIFT gateway and tracker, the Fedwire gateway, the core banking system, the general ledger and the bank's case tool — copying the UETR from one to the next. Free-text MT 199s and emails still arrive beside structured ISO 20022 messages, and the client or correspondent sets the deadline: “reply by 17:00 Frankfurt time”.
The dangerous replies are the ones that admit a mistake or move money. A reply that says the bank credited the wrong account accepts liability under UCC Article 4A, and the correction debits another customer; today the four-eyes check is often an email with a screenshot. A missed deadline turns into a fee claim, an interest claim or a complaint. The same correspondent sends the same problem again and again, and nobody tells the relationship manager. And SWIFT moves every investigation to camt.110 and camt.111 through Case Management by November 2027.
Estimated split of the people's working time on one non-receipt case, by hand and with the solution. It leaves out the waiting in between, which is why a case worked by hand takes days.
Seven agents read, trace, judge and draft every investigation; the analyst sends routine replies and a team lead approves anything that admits liability or moves money.
Thursday, October 22, 2026, at Corvane National Bank. On Monday, Halberd Bank AG in Frankfurt sent USD 48,250.00 for Feldmann Präzisionsteile GmbH's invoice INV-7741 to Marlowe Tool & Die LLC. At 8:52 a.m. Halberd's camt.110 arrives: the beneficiary says it was never credited, and the reply is due by 11:00 a.m. ET. Here is what happens next, screen by screen, in the working solution.
The Reading agent reads Halberd's camt.110, matches the UETR and notes the reply is due by 17:00 Frankfurt, 11:00 a.m. ET. The Research agent finds the pacs.008 in the payments hub, the SWIFT tracker's credited status (ACCC) at 10:42 a.m. on Oct 19, and the core posting: credited by account number to …4417, Marlow Tool Supply Inc. Marlowe Tool & Die LLC is …4471. Dana Whitfield's inbox, sorted by the client's deadline, puts the case at the top.
Research agent: “Beneficiary name check scored the names 0.81 against the 0.80 auto-pass threshold, and passed them with no review.”
Halberd follows up with a camt.056 cancellation: “If the beneficiary cannot be credited, please return the funds.” The Reading agent links it to the case by the same UETR. The Rules agent finds that the name and the number belong to different customers and Corvane's own check measured the difference, citing UCC 4A-207, Regulation J and the correspondent agreement. The Research agent confirms Marlow Tool Supply's balance of $212,904.16 covers the reversal, and §14.3 of its account agreement allows correcting a posting error, with notice.
Rules agent: “Correct the posting; do not return. The reply admits a Corvane-side miss.”
A camt.111 to Halberd: corrected with value Oct 19, Marlowe Tool & Die LLC credited USD 48,250.00, no charges. A camt.029 answering the cancellation: rejected, because the payment is applied to the named beneficiary. The correction entries: debit …4417, credit …4471 back-valued to Oct 19, no net effect on the general ledger. Then a notice to Marlow Tool Supply and a credit advice to Marlowe Tool & Die, from templates camt.111 v3.2, camt.029 v2.4 and notice v4.0, at a confidence of 0.94.
“This set admits liability and moves money, so a second person approves it. You cannot send it yourself.”
Dana opened the case at 9:14. Before she decides, she asks the agents: “Why correct instead of returning the funds?” The Rules agent answers in the thread: Halberd asked for a return only if the beneficiary cannot be credited, and it can, because Marlowe Tool & Die LLC holds …4471. A return would send Feldmann's money back and leave invoice INV-7741 unpaid.
Also suggested under the drafts: “Is any interest owed to Halberd?” and “Has Marlow Tool Supply used the funds?”
Because the set admits liability and moves money, the button reads Send for checking, with Edit and Reject proposal beside it. Dana sends it to Marcus Oyelaran, the team lead on duty, with a note: “Name check passed 0.81 on a different customer. Correct, do not return; both replies and both notices attached. Due 11:00 a.m.” In the checker queue, signed in as Dana, Approve is refused.
“You made this draft. Separation of duties: a second person approves it.”
Marcus sees the same thread, the same citations and the correction entries the maker saw. The amount is inside his $250,000 limit, so he approves with the reason “Evidence complete”. Approving posts correction CR-26-1022-0419, sends the camt.111 and camt.029 through SWIFT Case Management, and sends both notices. Had he returned it, he would have picked a reason code such as RC-03 Liability wording.
The Correction agent posts the entries: …4417 debited and …4471 credited USD 48,250.00, with value Oct 19. Halberd acknowledges both replies at 9:32 a.m., and the notice and the credit advice go out. No interest is owed, because the credit is back-valued. The outcome reads signed by Dana Whitfield and Marcus Oyelaran.
“Closed at 9:32 a.m. · 40 minutes after the request”
The case is Halberd's sixth name and account number mismatch in 90 days: 14 cases in all, the six worth $412,780. The Trends agent drafts a review note for the Nov 5 relationship review, with the cases behind each number. It also proposes raising the name-check auto-pass threshold to 0.90, for Grace Liu, the administrator, and Renata Kowalczyk, Head of E&I, to approve.
Trends agent: in the last 90 days, 0.90 “would have stopped 6 of 6 mismatches and added about 41 reviews a day”.
Every action by people and agents is on the trail with when, who, case, what, why and the evidence, and opening a source is logged too. Export packs are built for OCC examiners, internal audit, the BSA officer and a correspondent's request, with account numbers masked and Financial Crimes referrals only in the BSA pack.
The same Halberd case, seen by the four people around it — what their day looked like, and what it looks like now.
Seven agents read, trace, judge and draft every investigation; the analyst sends routine replies and a team lead approves anything that admits liability or moves money.
Reads each inbound request, structured (camt.110, camt.056, Fedwire camt.056) or free text (MT 199, email), classifies the case and extracts the UETR, amount, parties and the client's deadline, with a confidence on each field.
Finds the payment and its story: the payments hub record, the tracker status, the Fedwire leg, the core posting, the beneficiary check log and balances, citing each record.
Works out what the facts mean under UCC Article 4A, Regulation J, the SWIFT rulebook and the correspondent agreement, marks drafts that admit liability, move money or pay a claim, and computes compensation.
Writes the camt.111, camt.029, MT 199, client emails and customer notices from versioned templates, and removes restricted wording from anything a client or correspondent sees.
Drafts reversal, credit, return (pacs.004) and claim payment entries, value-dated, with the general ledger effect.
Watches every client deadline, predicts the cases at risk and why, escalates, and proposes how to rebalance work across the 22 analysts.
Summarizes repeat issues by correspondent and by cause, drafts relationship review notes, and proposes control changes such as a name-check threshold.
Approves what admits liability or moves money. The agents propose; a named person decides.
The analyst can question the agents in the thread before she decides, and the assistant answers about any case, payment, deadline or correspondent her role may see, with the message, posting and rule cited.
Why correct instead of returning the funds?
Halberd's camt.056 asks for a return only if the beneficiary cannot be credited. It can: Marlowe Tool & Die LLC holds …4471. A return would send Feldmann's money back and leave invoice INV-7741 unpaid. Correcting puts the funds where Halberd's instruction named, with value Oct 19, as the agreement asks.
Why didn't the name check stop the Halberd payment?
It scored “Marlowe Tool & Die LLC” against “Marlow Tool Supply Inc.” at 0.81, and the auto-pass threshold is 0.80, so it posted with no review. Six Halberd payments in 90 days had the same kind of mismatch. A proposal to raise the threshold to 0.90 is waiting for Grace Liu and Renata Kowalczyk.
Which cases will miss their client deadline today?
Six are at risk. Two are yours: PI-26-33795, where the SWIFT tracker did not answer, and PI-26-33802, waiting for Aldermoor Foods' consent. PI-26-33733 needs Renata Kowalczyk, who is in a meeting until 11:30 a.m.
Is any interest owed to Halberd?
No. The credit to Marlowe is back-valued to Oct 19, so no one loses the use of the funds. If Halberd claimed anyway, 3 days at the 3.95% federal funds rate would be $15.88.
13 screens from the working solution, on its sample data. Pick one to see it large.
The administrator sets them; every change is saved as a new version with who made it and why.
| Setting | Default | Choose from |
|---|---|---|
| Beneficiary name check · auto-pass threshold | 0.80 | A versioned rule; a change needs the administrator and the Head of E&I (0.90 proposed after the Halberd case) |
| What counts as “admits liability” | Needs a checker | Locked on: any reply saying the bank misapplied, delayed or wrongly charged a payment |
| What counts as “moves money” | Needs a checker | Locked on: any posting, reversal, return (pacs.004) or claim payment |
| Team lead approval limit | $250,000 | $100,000 · $250,000 · $500,000 |
| Claim payment limit | $10,000 | $5,000 · $10,000 · $25,000 |
| Default client deadline | One business day | Same day · One business day · Two business days |
| Escalation | Team lead 2 hours before; Head of E&I at the deadline | 1 hour · 2 hours · 4 hours before |
| Interest compensation | Federal funds rate, 360-day year | UCC 4A-506, from the reference-rate feed |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
“demo” = seen in the working solution, on its sample data · “target” = the design goal, measured in the live solution · “estimated” = our estimate · “typical” = published figures (SWIFT, “It's time to transform exceptions and investigations”, April 2025 — 5–10 days on average to resolve a payment investigation; an industry study of failed payments (2023) — $12.10 average fee per rejected or repaired payment; SWIFT's ISO 20022 exceptions and investigations timeline — camt.110 and camt.111 through Case Management by November 2027). Banks, companies, people, account numbers and references named on this page are fictional — characters and sample data in the working solution.
Payment Investigations is software in which payment operations teams work wire inquiries, cancellations and non-receipt claims as cases, with replies. Each case reads as one thread: the request, the agents' research steps with the record each one read, and the drafted replies. The analyst sends routine replies; a team lead approves any reply that admits liability, moves money or pays a claim.
The Reading agent reads the request and extracts the UETR, amount, parties and deadline. The Research agent finds the payment in the payments hub, the tracker, the core banking system and the general ledger. The Rules agent works out what the facts mean under UCC Article 4A, Regulation J and the correspondent agreement, and the Drafting and Correction agents write the replies, notices and entries for a person to send or send for checking.
Yes. A camt.110 investigation request is answered with a camt.111, and a camt.056 cancellation request with a camt.029; when a correspondent sends both, both replies are drafted in the same thread. Replies travel through SWIFT Case Management. Fedwire requests for return are worked the same way under Regulation J.
Yes. Paste a free-text request and the Reading agent shows each field it extracted with a confidence. Below 0.85 it asks rather than guesses: in the sample, an email date written “21/10” is read as October 21 at 0.78, and the analyst confirms it before the research runs.
It does not decide; it proposes. The Rules agent reads UCC 4A-207 and 4A-506, Regulation J, the SWIFT rulebook and the correspondent agreement, explains its finding with citations, and marks each draft that admits liability, moves money or pays a claim. The analyst can ask it why, edit a draft with an override reason, or reject the proposal with a reason code.
Yes. Agents never post, never send a reply that admits liability, and never contact a correspondent or client outside an approved reply. The analyst sends routine replies; anything that admits liability, moves money, pays a claim or sends a customer notice is approved by a team lead, the maker can never approve their own draft, and larger amounts go to the Head of E&I.
No. Financial Crimes referrals are restricted and never shown on a client or correspondent view or in a reply. A client service officer sees status only, and the Drafting agent removes restricted wording from every client and correspondent reply. Referrals appear only in the BSA officer's export pack.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your correspondent agreements, reply templates, approval limits and a sample of past investigations — and it goes live once every quality gate has passed. We will walk you through it on your own cases first.
Every failed payment fixed with its proof, checked by a second person, before the cutoff.
1-min demoACH Returns & NOC ManagementEvery return inside its window, and every originator caught before it crosses a level.
1-min demoISO 20022 Structured Address RemediationEvery client and payee address structured, proven and signed twice before 14 November.
1-min demoWe'll run Payment Investigations on a sample of your own past wire inquiries and cancellations.