Trade-based money laundering, shown as a price and a route before anyone reads a document
Every trade alert decided in days, with the evidence that decided it.
See one case, screen by screen ↓Trade-based money laundering detection is how a bank finds value being moved through its trade business — over- and under-invoicing, phantom or circular trades, routes that make no commercial sense, and payers with no recorded relationship to the trade. This solution screens every presentation, documentary collection and reimbursement against 86 cited red flags, then has six agents reason across price, route, parties and documents and write the case, with a named person deciding every outcome.
AI use cases in banking: a buyer's guide to agentic AI for bank and credit union operations →
Updated
A bank sees the documents, never the goods. An invoice for polyester filament yarn at USD 18.40 a kilogram looks exactly like one at USD 2.40 unless somebody knows what that yarn costs. FinCEN said as much when it issued its trade-based money laundering advisory in 2010, and the FFIEC manual still tells examiners to ask whether the bank reviews trade documents for anomalies and red flags, not only for compliance with the credit. Trade finance AML is therefore a reasoning problem: is this price plausible for these goods, does this route make commercial sense, and do these parties keep paying each other in a circle?
Today that reasoning is manual. An investigator looks up a trade statistic, scrolls a shipping site, searches the bank's own systems for the parties, and then types a paragraph — a median of 19 days against a 30-day clock. And the risky half is the clearances, because an examiner samples the alerts that were closed. A closure with no written reason and no evidence behind it is the finding, however correct the judgment was.
Estimated split of the 19-day median from alert to a decided outcome at the demo bank, by hand and with the solution. The 30-day clock is the bank's own policy, matched to the reporting clock.
Six agents check the price, the route, the parties and the documents, then argue the innocent explanation; two named people decide.
Corvane National Bank holds import letter of credit ILC-2026-004871 for USD 2,208,000 in favor of Lanmere Textiles Pte Ltd, on the account of Brightwater Textile Supply LLC in Houston: 120,000 kg of polyester filament yarn, HS 5402.33, at USD 18.40 a kilogram. Here is what happens next, screen by screen, in the working solution.
The screening reads 44 presentations and collections and 61 reimbursements against 86 trade red flags, raises 10 alerts, and the Challenger agent clears 7 of them with a written explanation and the document that proved it. Three reach a person's queue, each with its own 30-day clock running from the date of detection.
The queue also says what it could not confirm: “Vessel tracking data has not refreshed since 04:10. Route checks on 4 alerts are marked unconfirmed, and no alert has been closed on a route flag since.”
Alert TBM-2026-0418 opens on the shipment itself, not on a list. A route strip runs from Ningbo through Jebel Ali to Houston with the paying and paid bank above each leg, and below it the unit price sits as a single dot on the 12-month market band for this goods code, with the customer's own nine shipments as gray ticks inside the band.
“47 days in all — the direct Ningbo to Houston service sails weekly and takes 30, so this voyage adds 17 days through a free-trade zone.”
The band for HS 5402.33 runs from USD 2.10 to USD 3.60 a kilogram, median USD 2.74, from 4,186 observations. USD 18.40 is 5.1 times the top of that band and 6.7 times the median; the customer's own nine shipments of the same code over 18 months ran from USD 2.35 to USD 3.05. The flag names FinCEN's over-invoicing red flag and the two exhibits behind it.
Clicking the band opens the market price band exhibit at the highlighted passage: the goods code, the 4,186 observations, the 5th percentile, the median and the 95th percentile, and the USD 18.40 declared on this presentation. Nothing on that view can be edited, and one control adds the exhibit to the evidence pack.
The Document agent compared every field of the credit, both amendments and all six presented documents and found three mismatches. The Pricing agent built the band and the customer's history. The Route agent pulled both voyage records. The Party agent linked five parties across 23 transactions. Each panel shows what it read, what it found, its confidence and who it handed to.
The Route agent says why it stopped at 81%: “the Jebel Ali berth record is from the carrier, not the port.”
The USD 2,208,000 reimbursement reached the bank from Velmara Holdings FZE in the Jebel Ali Free Zone, a party the credit never names. Over the previous 11 months the beneficiary sent that same party USD 1,960,000 in four wires, and the notify party on nine of eleven presentations shares its registered address. Every link opens the transactions behind it.
Three innocent explanations are written and tested. A specialty high-tenacity grade would carry a premium — but the mill certificate and specification sheet are not in the file, and the packing list and certificate of origin both describe the ordinary 150D/48F semi-dull. A landed, duty-paid price would explain part of it — but the invoice is FOB Ningbo. A group-treasury payer would explain the money — but no affiliate outside Singapore is recorded.
The agent may never clear a flag on the customer's say-so, and never when the proving document is absent.
The Case writer recommends escalation under reason code TBML-OVP-01 and shows the clock: detected October 6, 2026 at 23:52, so a decision is due November 5. Marisol chooses to escalate, picks the reason code, writes her own sentence for the record, and signs.
The maker card reads “Signed Oct 9, 16:40 · TBML-OVP-01”; the checker card waits for the BSA Officer. The approve control is unavailable to the person who proposed it, with the reason beside it. Below, everything that has happened to this alert sits on one trail — the agents' findings, the document requested from trade operations, the evidence pack previewed, the proposal signed.
“You proposed this escalation. Approval is the BSA Officer's step — nobody approves their own.”
The case narrative runs from the summary through the transaction, price, route, parties and payment, documents and the explanations considered, and every sentence carries a numbered exhibit that opens at its passage. A banner records that the case, and the fact that it exists, is not shown to trade operations, to relationship staff or on any customer-facing view, and that exports are redacted by rule.
The same alert, seen by the four people it passes — what their month looked like, and what it looks like now.
Six agents check the price, the route, the parties and the documents, then argue the innocent explanation; two named people decide.
Normalizes the declared price to a price per kilogram, builds the 12-month market band for the goods code from public trade statistics, and plots the customer's own 18 months against it. Reports the multiple of the band top and of the median, with the observation count.
Reads the bill of lading, pulls each voyage record, draws the route from the port of loading through every transshipment to discharge, and tests the total against the fastest scheduled direct service on that corridor and against the free-trade-zone and high-risk port lists.
Links applicant, beneficiary, shipper, notify party and payer across 18 months of letters of credit, documentary collections and wires the bank already holds, and looks for third-party payers, value returning between the same parties and shared addresses.
Reads the credit, every amendment, the invoice, the bill of lading, the packing list and the certificate of origin, and compares description, grade, quantity, weight, parties, ports and dates field by field, listing each mismatch with both passages and each amendment with the reason given.
Writes the most likely innocent explanation for every flag, names the exact document that would prove it, searches the file for that document, and clears the flag when it is there — recording both the explanation and the proof.
Assembles the case narrative in the unit's template — summary, the transaction, price, route, parties and payment, documents, the explanations considered and the recommendation — with a numbered exhibit on every sentence, and drafts the escalation memo.
Decides; the BSA Officer approves and files. The agents propose; a named person decides.
Ask the assistant in plain words — or tell it what to change. Every answer points at the exhibit it came from.
Why is this price a red flag?
USD 18.40 a kilogram is 5.1 times the top of the band for this goods code. The 12-month band for HS 5402.33 is USD 2.10 to 3.60, median 2.74, from 4,186 observations, and the customer's own nine shipments of the same code over 18 months ran USD 2.35 to 3.05 — so it is out of character for them as well as for the market. The only reading that would explain it is a specialty grade, and the packing list and certificate of origin both describe the ordinary 150D/48F semi-dull.
Which alerts could miss the 30-day clock?
Three. TBM-2026-0384, Tallowbridge Chemicals Inc, is on day 27 of 30 and its escalation summary is drafted. TBM-2026-0388, Orrin Cold Chain LLC, is on day 24 and must be with the BSA Officer tomorrow. TBM-2026-0391, Sutterfield Auto Parts LLC, is on day 22 and the second bank has not answered the request. The queue is filtered to them.
Could a specialty grade explain the price?
No, and two documents say so. The invoice calls it a specialty high-tenacity grade, but packing list PL-26-3318 and certificate of origin CO-SG-26-1140 both describe 150D/48F semi-dull, the ordinary grade. A mill certificate and a specification sheet would settle it, and neither is in the file. Even a true specialty grade trades within about 1.6 times the band top on this code.
Raise the price alert threshold to three times the band
Proposed, not applied. Over the last 12 months, moving that rule from 2.0× to 3.0× would have raised 388 fewer flags and would have missed 2 of the 61 escalations that became cases. A threshold change needs the BSA Officer's approval, so it has been sent to her with those two cases attached.
13 screens from the working solution, on its sample data. Pick one to see it large.
The red flags come from FinCEN, the FFIEC manual, FATF and the Wolfsberg Group and are cited on screen; what they are set to is yours, versioned, and approved by a named person before it takes effect.
| Setting | Default | Choose from |
|---|---|---|
| Band window | 12 months | 12 · 18 · 24 months |
| Minimum observations before a band is used | 40 observations | 40 · 60 · 100 observations |
| Raise a flag above | 2.0× the band top | 2.0 · 2.4 · 3.0× the band top |
| Recommend escalation above | 5.0× the band top | 4.0 · 5.0 · 6.0× the band top |
| Transshipment days over the direct service | 10 days | 7 · 10 · 14 days |
| Party look-back | 18 months | 18 · 24 · 36 months |
| Confidence floor | 65% | 60 · 65 · 70% |
| With the BSA Officer by | Day 25 | Day 22 · 25 · 27 |
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 trade data · “target” = the design goal, measured in the live solution · “estimated” = our estimate from the demo bank's design volumes · “typical” = published figures (FinCEN Advisory FIN-2010-A001, February 18 2010 — over 17,000 suspicious activity reports describing trade-based money laundering between January 2004 and May 2009, covering more than $276 billion; 31 CFR 1020.320(b)(3) — a report no later than 30 calendar days after the date of initial detection, 60 where no suspect is identified; 31 CFR 1020.320(e) on confidentiality; FFIEC BSA/AML Examination Manual, Trade Finance Activities; FATF trade-based money laundering risk indicators; Wolfsberg Group, ICC and BAFT Trade Finance Principles). People, companies, banks, vessels and references named on this page are fictional — characters and sample data in the working solution.
Trade-based money laundering detection is how a bank finds value being moved through its trade business — over- and under-invoicing, phantom or circular trades, routes that make no commercial sense, and payers with no recorded relationship to the trade. This solution screens every presentation, documentary collection and reimbursement against 86 cited red flags, then has six agents reason across price, route, parties and documents and write the case, with a named person deciding every outcome.
The library on screen carries 86 of them, each cited to FinCEN's trade-based money laundering advisory, the FFIEC BSA/AML manual's trade finance section, FATF's risk indicators or the Wolfsberg, ICC and BAFT Trade Finance Principles — among them a unit price outside the market band, a third-party payer the credit does not name, an amendment that widens the goods description with no reason, a transshipment beyond the direct service, and goods described differently across documents.
The Pricing agent normalizes the declared price to a price per kilogram, builds a 12-month band for the goods code from public trade statistics, and plots the customer's own shipments of that code against it. On the sample alert, USD 18.40 a kilogram sits 5.1 times above a band of USD 2.10 to 3.60 while the customer's own nine shipments all sit inside it. The agent never scores a code with fewer than 40 observations; it says there is no market data and asks for an analyst.
No. Examining a presentation under UCP 600 is trade operations' decision, on their own five-banking-day clock, and the Document agent is explicitly barred from re-examining it. The financial crime case runs alongside and is never visible on that screen.
No. As 31 CFR 1020.320(e) requires, no case, alert or filing indicator appears on a trade operations screen, a relationship screen or any customer-facing view, and exports are redacted by rule. In the working solution you can switch to the trade operations role and open the same shipment: the credit, the documents and the clock are there, and nothing else is.
They close with the innocent explanation the Challenger agent wrote and the document that proved it, both kept for five years and both visible to an auditor. That matters because an examination samples the closed alerts first: a clearance with a written reason and its proof reads as a decision, not a dismissal.
Yes. Agents read, compare, draft and propose; they never close, escalate or file. The investigator decides, the BSA Officer approves every escalation and decides and files every report, and neither can take the other's step. Every closure carries a reason code, the evidence considered and a signature whose meaning is recorded.
The Agentic Solution Engine builds and deploys it from your requirements, your documents and your rules — the typologies and their thresholds, the clocks, the reason codes, the approvers and the roles — and it goes live once every quality gate has passed. The red flags ship cited; what they are set to, and who signs what, is configured to your policy.
Every party, bank, vessel, port and goods line screened, cited and decided the same day.
1-min demoTrade Finance Fraud DetectionEvery cargo fingerprinted and checked against the bank's own financing before funding.
1-min demoUCP 600 & ICC Rules SearchEvery discrepancy question gets a cited verdict in seconds, the reasoning kept for audit.
1-min demoWe'll run Trade-Based Money Laundering Detection on a sample of your own past presentations and alerts.