SSI management where every change to a standard settlement instruction is proven at the source and called back before a dollar moves
Every SSI change proven at the source and called back before a dollar moves.
See one case, screen by screen ↓SSI management is the work of keeping counterparty standard settlement instructions current and proving every change before a payment uses it. Standard Settlement Instruction Management does it with seven agents that read each change request in any language, score it for fraud, find the counterparty's own record and book the call-back to the number on file, and lay it all out in one dossier you can question in plain words. An analyst calls and verifies, a different team lead approves or freezes, and financial crime decides the cases.
AI in asset management: a buyer's guide to agentic AI for investment managers →
Updated
SSI management is the work of keeping every standard settlement instruction current — the bank or custodian, account, place of settlement and intermediary a counterparty settles to — and of proving each change before a payment uses it. Quillfield Capital, the asset manager in the working solution, holds 18,412 live instructions for 1,146 counterparties in 38 currencies, used by 412 funds and mandates, and receives about 600 change requests a month: industry SSI database notices, emails, signed letters in the counterparty's own language, portal uploads and Swift MT 671 messages. Each one is retyped into the SSI master, called back on a number someone has to dig out, and approved by a second person by hand.
The risk is the request that looks like every other row in the inbox. Payment instruction fraud works because a forged change is well formed, passes every BIC and IBAN check and can come from a domain one hyphen away from the real one — and the phone number in the email reaches the fraudster. A dormant or outdated instruction fails trades instead, and under T+1 settlement there is no second day to fix it. When an auditor asks who changed an account, on what proof and who checked, the answer is spread across a mailbox, a phone log, a spreadsheet and the master.
Estimated split for one bank or account change by email or letter, by hand and with the solution. The call-back itself stays with a person.
Seven agents read, score, find the counterparty's own record and book the call-back; an analyst verifies, a different lead approves.
Monday, October 19, 2026, 23:10 London. An email from “Simon Hale” at brennmoor-sec.com asks Quillfield Capital to send all of Brennmoor Securities' US dollar cash settlements to a new account at Pafos Merchant Bank in Limassol, Cyprus, “effective immediately.” 23 trades worth $41.6 million settle to Brennmoor's instruction on October 20–21. Here is what happens next, screen by screen, in the working solution.
Marcus's home shows four numbers: 47 open change requests with 16 due today, 3 held by the agents, 9 waiting for a checker and 26 going live this week. Under “Needs you” there is one next step each, Brennmoor first: “a USD account change the counterparty denies on the call-back.” Beside it the agents report the night — 31 requests read from 5 channels and 12 call-backs booked, all to numbers on file — and below it a box to ask them, with “What needs me before the 10:00 payment run?” one click away.
“Held 3: Brennmoor 92, Oakharrow 88, Ribeira do Sul 71” — Fraud agent, 23:14
Every request is read, linked to the instruction on file and scored before a person sees it: 171 industry database notices, 142 emails, 58 signed letters, 31 portal uploads and 16 Swift MT 671 messages this month. “Held by the agents” shows three: Brennmoor at 92, Oakharrow FX at 88 because its new RUB bank is on the sanctions list, and Banco Ribeira do Sul at 71. Oakharrow has already escalated to Robert Haines with 82 minutes left on its 4-hour clock.
“SLA clocks: high risk 4 hours, standard 1 business day, authenticated 2 business days.”
The Intake agent read the email and its one-page letter and mapped six changed fields on SSI-04417 — account, bank, BIC, bank country from the US to Cyprus, a new intermediary and the ABA routing number removed — each value cited to its line. The Fraud agent scored it 92: a lookalike sender domain registered October 13 (+30), a new bank country with no Brennmoor entity there in its LEI record (+20), a signer not on the authorized list (+10) and four signals of +8. The Source agent found Brennmoor's own authenticated record unchanged since June 2, and the Checker agent passed the BIC, IBAN and sanctions tests, then said what that does not prove.
“Well formed is not the same as genuine.” — Checker agent
The Fraud agent refused +357 25 030 418 because it arrived with the request. The Callback agent booked the call to +44 20 7946 0381, Brennmoor's settlements desk on the master record, re-verified February 11, and wrote the script: ask them to read their details out, never read the new ones to them. Aisha calls on the recorded line; 2 min 41 s later the transcript is on the dossier and the Callback agent has compared it field by field: 0 of 6 changed fields confirmed.
Gemma Ward, Brennmoor's settlements manager: “No. We haven't changed our USD instructions, and we sent no such request. There's no Simon Hale in settlements.”
Under the three panels the dossier shows how the agents worked — every step in order, with its time, ready to replay — and the conversation with them, each answer with its confidence and sources. Marcus asks: “What else did this sender try?” The Fraud agent answers that two earlier requests this month asked for the same Cyprus bank, PFMBCY2L, from newly registered lookalike domains: Ostervik Capital Markets on October 7 and Kaldera Bank on October 15. Both were frozen.
“This looks like one campaign; I will link all three to the case when you freeze.” — Fraud agent
With the call-back denied, the dossier offers no Approve, only Freeze and escalate or Reject. Marcus chooses Freeze and escalate, keeps reason code FR-01 — “Counterparty denied the request on a call-back to the number on file” — and signs. SSI-04417 is locked against email and letter changes for 90 days, financial crime gets a case with the evidence, and the agents propose blocking the sender domain. The 23 trades settle to Halden Trust as before.
“I, Marcus Lee (SSI Team Lead), sign this decision. My name, the time and the evidence are added to the audit trail.”
Case FC-2026-0219 opens with the evidence already gathered: the email as received, headers showing SPF and DKIM pass for the lookalike domain, the domain registered October 13, the call-back transcript and Brennmoor's unchanged record. The Fraud agent's pattern links it to Kaldera Bank ($12.9M protected) and Ostervik Capital Markets ($7.4M): three requests in 12 days, one bank, domains registered 5 to 9 days before use, all sent after 21:00 London. Telling Brennmoor on the number on file, blocking the domain, adding the bank to the watch list and the report decision each need a financial crime officer and are logged.
Fraud cases are “decided by financial crime, never by the agents.”
Kawabe Securities' signed letter moves its JPY equities settlement agent from Sakuragawa Trust Bank to Minato Custody Bank from October 26. The Intake agent shows the Japanese beside the English, every field cited to its line; the Source agent finds the same change in Kawabe's own authenticated record, entered October 16; the risk is 14. Wei Ling's call-back to Misaki Sato on the number on file confirms all three changed fields, and she verifies as maker. Marcus approves as checker with AP-01; had he verified it himself, the solution would not let him approve it.
“Minato Custody Bank, account 0081-447120, from October 26.” — Misaki Sato, on the call-back
The approved change is scheduled for 00:01 Tokyo on October 26 in the SSI master, the order management system, the custodian and the payment platform, and the Go-live agent collects a receipt from each; trades settling before then stay on Sakuragawa Trust Bank. The calendar shows 27 changes for October 20–26. When custodian B's link failed at 07:58, the banner said so — 3 messages queued, a retry every 10 minutes — with Retry now one click away.
The Go-live agent never activates a change without a maker and a different checker, and never moves an effective date.
The audit trail records who, what, when, why and on which evidence: Aisha's call-back at 08:12, the Callback agent's comparison at 08:16, Marcus's freeze with FR-01, the Fraud agent opening the case and locking SSI-04417, Wei Ling's verification and Marcus's approval. It is append-only, hash-chained and kept 7 years. Verify chain checks it, and Export pack for auditors builds the evidence for one request, one instruction or an exam request.
“I need to prove every instruction we paid to was ours to pay.” — Robert Haines
The same forged request and the same genuine one, seen by five of the people who carry them — what the month looked like with a mailbox and a phone log, and what it looks like now.
Seven agents read, score, find the counterparty's own record and book the call-back; an analyst verifies, a different lead approves.
Reads change requests from email, signed letters, portal uploads, industry SSI database notices and Swift MT 670/671 messages in any language; maps every field to the FMSB/ISITC template with a citation to its line, shows foreign-language originals beside the English, links each request to the counterparty and instruction on file, merges duplicates and drafts a request for missing fields to the contact on the master record.
Scores every request from 0 to 100 with weighted, explained signals — lookalike domain detection with the domain’s registration date, a new bank country against the LEI record, signer authority, business hours, urgency, disagreement with the counterparty's own record — holds it at 70 and links attempts across counterparties for financial crime.
Finds the counterparty's own authenticated record in the industry SSI database and its signed letters, cites them, and flags any disagreement with the request.
Tests each new instruction on its own, apart from the fraud score: BIC, IBAN and ABA checks, account format by market, place of settlement, sanctions and watch lists, the account-name check and FMSB template completeness.
Books each call-back to the number on the master record inside the counterparty's business hours, writes the script, transcribes the recorded call and compares each answer with the changed fields, then suggests an outcome for the caller to decide.
Schedules approved changes at the agreed time, pushes them to the SSI master, order management, custodians and the payment platform, collects a receipt from each, retries failed pushes and prepares rollbacks.
Builds campaigns for dormant (12 months unused), failed-trade and high-risk instructions, compares each with the counterparty's own record and its usage, and proposes confirm, retire or call back — 688, 431 and 167 in the sample Q4 campaign of 1,286.
Maker, checker, cases, evidence. The agents propose; a named person decides.
Every screen has a box to ask or instruct the agents: today's work, the queue, each request, each call-back, the approvals, the SSI master, the go-live calendar, recertification, fraud cases, the audit trail and the settings. Answers show their confidence and cite their sources; anything that changes an instruction comes back to a maker and a different checker to sign.
What needs me before the 10:00 payment run?
Two things. Brennmoor (risk 92): Brennmoor denied the request on the call-back; freezing it keeps $41.6M on the instruction on file. Oakharrow: the new bank is sanctioned; reject it. Nothing else touches today's run.
What else did this sender try?
Two earlier requests this month asked for the same Cyprus bank, PFMBCY2L, from newly registered lookalike domains: Ostervik Capital Markets on Oct 7 (5 days old) and Kaldera Bank on Oct 15 (9 days old). Both were frozen. This looks like one campaign; I will link all three to the case when you freeze.
Does the English match the Japanese?
Yes, line by line. The date 2026年10月26日 is Monday, Oct 26; 港カストディ銀行 is Minato Custody Bank; the account 0081-447120 and BIC MNTCJPJT are copied as written.
Select the ones I can approve together
Selected 5: each has its source, the counterparty's own record or a confirmed call-back, passing checks, risk under 30 and a maker who is not you. You sign once, with one reason code; each goes live at its own agreed time.
13 screens from the working solution, on its sample data. Pick one to see it large.
The rules every request is checked against are settings. A change is a proposal, tested on last month's requests and approved by a second person — a different administrator or the Head of Settlement Operations. On September's 597 requests, a hold at 70 would have held 9, including all 3 confirmed fraud attempts.
| Setting | Default | Choose from |
|---|---|---|
| Hold a request at a fraud score of | 70 | 40 to 90, in steps of 5, with the impact on last month shown |
| Call-back required for | Every bank, account or intermediary change; every email or letter; any request scoring 30+ | On |
| Never use contact details that arrive with a request | Locked on | Cannot be switched off |
| Minimum notice before go-live | 1 business day; 2 for a bank-country change | A proposal a second person approves |
| Lock after a confirmed fraud attempt | 90 days: email and letter changes refused | A proposal a second person approves |
| Dormant after | 12 months unused, or any failed trade (FMSB Core Principle 7) | A proposal a second person approves |
| Fraud signal weights | Lookalike domain +30 · new bank country +20 · signer not authorized +10 · sanctioned bank +60 | Contact number with the request, outside hours, urgency, disagrees with own record: +8 each |
| SLA clocks | High risk 4 hours · standard 1 business day · authenticated 2 business days | High risk escalates to the Head of Settlement Operations with 2 hours left, and again on breach |
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 (TABB Group research for DTCC — about 20% of trade fails from incorrect standing settlement instructions; FBI IC3 2024 Internet Crime Report — $2.77 billion of business email compromise losses). Guidance named on this page: FINRA Regulatory Notice 12-05; FMSB Standard for Sharing of Standard Settlement Instructions V2, Core Principle 7; SEC Rule 204-2. Quillfield Capital and the people, counterparties, banks, domains, phone numbers and accounts named on this page are fictional — characters and sample data in the working solution.
SSI management is the work of keeping counterparty standard settlement instructions current and proving every change before a payment uses it. Standard Settlement Instruction Management does it with seven agents that read each change request in any language, score it for fraud, find the counterparty's own record and book the call-back to the number on file, and lay it all out in one dossier you can question in plain words. An analyst calls and verifies, a different team lead approves or freezes, and financial crime decides the cases.
SSI validation runs three independent ways. The Checker agent tests the BIC, IBAN and ABA number, the account format by market, the place of settlement, sanctions and watch lists, the account-name check and FMSB template completeness. The Source agent compares the request with the counterparty's own authenticated record in the industry SSI database and its signed letters. And every bank, account or intermediary change, every email or letter and any request scoring 30 or more is called back on the number on file — because a forged request can pass every format check, as the one in the working solution did.
The Fraud agent scores every request with weighted, explained signals — a lookalike sender domain and its registration date, a new bank country, a signer not on the authorized list, a contact number sent with the request, out-of-hours timing, urgency and disagreement with the counterparty's own record — and holds anything at 70 or above for a team lead. It refuses any contact detail that came with the request and links attempts across counterparties, so financial crime sees a campaign rather than one email.
Callback verification means confirming a settlement instruction change by phoning the counterparty before it is used, and the number matters because whoever sent a forged change also answers the number in it. FINRA Regulatory Notice 12-05 makes the same point to broker-dealers about emailed instructions to move customer funds: confirm the customer really sent them, and treat urgency or an unfamiliar account as red flags. Here that rule is locked on: the Callback agent books only the number on the master record, inside the counterparty’s business hours, writes the script and transcribes the recorded call; a person makes the call and records the outcome with a reason code.
A trade enriched with an outdated or wrong instruction cannot match or settle, and published research puts incorrect SSIs behind about 20% of trade fails. Under T+1 settlement there is no second day to fix one, and Europe, the UK and Switzerland move to T+1 on October 11, 2027. The Recertification agent reviews instructions unused for 12 months or used in a failed trade, as FMSB Core Principle 7 suggests; the target is half of instruction-caused fails removed at source.
Yes. It reads change notices and authenticated records from the industry SSI database and Swift MT 670/671 messages alongside emails, signed letters and portal uploads, and pushes approved changes to the SSI master, the order management system, custodians and the payment platform with a receipt from each. Sanctions lists, the BIC directory, the LEI registry, domain registration lookup and recorded telephony are connected too.
No. The agents read, score, find, check, book and schedule; they never make or confirm a call-back, clear a hold, approve or retire an instruction, change the SSI master before a maker and a different checker sign, or move an effective date. People stay in control: an analyst verifies, a different team lead approves, anyone who can release payments cannot approve instructions, and financial crime decides every case.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your SSI policy, call-back procedure, an export of the SSI master and a sample of past change requests — and it goes live once every quality gate has passed. Setup connects the mailbox, the industry SSI database and Swift, imports the master, maps fields to the FMSB/ISITC template, re-verifies call-back numbers and replays past requests against your rules first.
Target 20–30% fewer fails, each fix cited and checked before the market’s cut-off.
1-min demoPosition and Cash ReconciliationEvery break explained from its source before the NAV is struck, every fix proved first.
1-min demoCorporate Actions ProcessingEvery event one cited golden record, every election made before the custodian’s cut-off.
1-min demoWe'll run Standard Settlement Instruction Management on a month of your own past change requests and the SSI master they changed.