Leaving this course?

Lesson 1 of 6

Why they grew up apart

Fraud and anti-money laundering sit in different parts of almost every institution, and the separation is not an accident or an oversight. It follows from what each was built to do.

Different purposes. Anti-money laundering is a regulatory obligation: identify customers, monitor activity, report suspicion to the state. Its output is intelligence handed to someone else. Fraud is loss prevention: stop money leaving, recover what left, keep the customer. Its output is a number on the profit and loss account.

Different owners and incentives. The money laundering reporting officer answers to a regulator and, ultimately, to criminal liability. The head of fraud answers to a loss budget. One is judged on whether reports are good; the other on whether losses are low. Neither metric rewards helping the other.

Different data, and this is the part that matters most. Fraud teams work with device fingerprints, session behaviour, typing cadence, IP reputation, velocity and card-present signals — data collected in seconds and often discarded in weeks. AML teams work with customer identity, source of funds, expected activity and transaction history over years. Fraud data is richer and shorter-lived; AML data is thinner and longer-lived. They were collected for different stated purposes, which has consequences covered in lesson three.

Different systems. Separate monitoring engines, separate alert queues, separate case managers, separate investigators, often separate reporting lines up to different executives. An institution can have two teams looking at the same payment from two systems that cannot see each other.

Different clocks. Fraud decides in milliseconds — authorise or decline, hold or release. AML decides over days — review the alert, gather the facts, form a suspicion, report. A function built for one tempo does not naturally serve the other.

What the separation costs. Three things, and they are measurable:

  • The same customer looks different to each team, and neither sees the whole person.
  • Fraud proceeds are criminal proceeds. A fraud stopped and never reported is a laundering typology that the financial intelligence unit never learns about.
  • Each team's action can destroy the other's work — the clearest example being an account closed for fraud that the AML investigation needed kept open, and that is before anyone mentions tipping off.

The honest counter-argument. Separation also has real benefits: focus, speed, and expertise that is genuinely different. "Convergence" done badly produces one team that is worse at both jobs. The aim is not to merge everything; it is to stop the wall running through places where there is only one problem.

Terminology: the FATF standards say suspicious transaction report (STR), while the United States and the United Kingdom say suspicious activity report (SAR). They are the same filing, and this course uses whichever term its own sources use.

Lesson 2 of 6

Where they are the same problem

Five places where fraud and money laundering are not two subjects but one, and the division is an artefact of the organisation chart.

Money mules. The clearest case. A fraud victim's money lands in someone's account and is moved on within minutes. To the fraud team that is the destination of a loss; to the AML team it is placement and layering by a third party. It is the same account, the same thirty minutes and the same evidence, and the receiving institution usually holds more of it than the sending one.

Authorised push payment scams. The payment was authorised by the customer, so fraud systems designed around unauthorised use may not stop it. The receiving side is an AML problem and the sending side a fraud problem, and the reimbursement question sits on top of both.

First-party fraud. A customer who lies on an application, or who borrows with no intention of repaying, is committing fraud. The proceeds are criminal proceeds, which makes their subsequent movement laundering. Many institutions treat first-party fraud purely as credit loss and never consider reporting it.

Insider fraud. An employee moving money, or selling access, is both a fraud and — because the proceeds must be banked — a laundering matter. It also sits inside the institution's own controls, which is why it is so often handled quietly by human resources and never reaches either function properly.

Fraud as a predicate offence. This is the systemic point, and it is the one the FATF, the Egmont Group and INTERPOL made in their 2023 work on cyber-enabled fraud: the proceeds are enormous, they are generated at scale across borders, and they have to be laundered like any other proceeds. Treating fraud as a loss event rather than as a predicate crime means a large share of the world's criminal proceeds never enters the anti-money laundering system at all.

The common structure underneath. In each case the two functions are looking at different ends of one chain: fraud sees the event, AML sees what happens to the money afterwards. Fraud has the detail and the speed; AML has the obligation, the history and the route to the state. Neither end is sufficient on its own, which is the whole argument for connecting them.

Lesson 3 of 6

The constraints nobody mentions

Most writing on convergence stops at "share more". These are the reasons it is harder than that, and they have to be designed around rather than wished away.

Tipping off. The sharpest problem, and the least discussed. A fraud team's ordinary work includes telephoning the customer: did you make this payment, is this your device, we are holding this transaction. If an anti-money laundering suspicion has been formed or reported on the same customer, that call can constitute tipping off — not because anyone mentions a report, but because the customer learns their activity is under examination. Three practical consequences:

  • Fraud outreach on a customer with a live AML suspicion needs a rule, decided in advance, about who may contact whom and what may be said.
  • A single case management system makes the risk greater, not smaller, because a fraud investigator can now see the AML file. Access has to be designed, not inherited.
  • "Do not contact" flags must exist and must be respected by the fraud function, which is used to contact being the first response.

Purpose limitation and legal basis. Fraud data is typically collected and processed on one legal basis and AML data on another, often a statutory obligation. Using fraud data for AML purposes, or the reverse, is a processing decision that needs a basis — not merely a technical integration. In jurisdictions with a data protection regime this is where a convergence project either gets designed properly or quietly becomes non-compliant.

Retention. Fraud signals are often kept for weeks or months; AML records for years, commonly five or more. Joining the two means deciding whether fraud data is now being retained for the longer period, and if so, under what basis and with what notice.

Different evidential standards. Fraud often acts on probability and acts fast, reversing later if wrong. Reporting a suspicion is a lower bar than proving a crime but a considered one. A converged alert queue mixing a millisecond decline with a week-long suspicion assessment will serve one badly.

Confidentiality in the other direction. An employee suspected of insider fraud has employment rights; a customer has data rights. Both constrain what can be shared internally, and neither disappears because the two teams now sit together.

And one that is cultural rather than legal. Fraud teams are measured on speed and on customers retained; AML teams on completeness and defensibility. Put them in one room without reconciling those measures and the louder metric wins — which, in most institutions, is the one with money attached.

Lesson 4 of 6

How to join them, from cheapest to deepest

Convergence is a spectrum, not a switch. Each step below is worth doing on its own, and most institutions should stop well before the end.

1. Liaison. A named person each side, a standing meeting, and an agreed route for "you should look at this". Costs nothing, requires no system change, and removes the most common failure — nobody told anyone.

2. Shared typologies. One list of patterns that matter to both, written jointly: mule shapes, scam typologies, first-party indicators, insider patterns. This is where the two vocabularies get reconciled, and doing it reveals how much each team already knew that the other did not.

3. Decision rules at the boundary. Written answers to the questions that cause damage: who may close an account and who must be consulted first; who may contact a customer when the other function has an open matter; who decides to hold a payment; and who files the report. The account-closure rule is the one to write first, because an account closed by fraud is an AML investigation ended by someone who did not know it existed.

4. Shared signals, not shared systems. Feed specific fraud signals into AML monitoring and specific AML context into fraud decisions, without merging the platforms: a mule-pattern flag, a device shared across customers, a known-scam beneficiary, a customer under AML review. Cheaper and far less risky than integration, and it captures most of the value.

5. A shared case view, with designed access. One place to see what the other function knows about a customer, with access controlled by role and the tipping-off rules built in rather than assumed.

6. A single financial crime function. One leader, one set of metrics, combined investigation. It works in institutions large enough to keep specialists inside it, and it fails where it becomes generalists doing both jobs adequately. The test is whether your best mule analyst and your best sanctions analyst are both still doing what they are good at.

What to share externally, too. Mule chains and scam networks cross institutions, which is why public-private partnerships and collaborative analytics have produced results that no single institution's convergence project could. Internal convergence makes an institution a better participant in those; it does not substitute for them.

Lesson 5 of 6

One model or two?

The sharing question has a technical twin that gets far less attention: should the two functions detect with one model or two? The signals should be shared and the decisions should not — and a good number of institutions get this exactly the wrong way round.

The objectives are different, and not by a little. A fraud model exists to minimise expected loss. Every alert it raises has a price — a declined genuine payment, an irritated customer, a call to the contact centre — weighed against money that would otherwise leave the building. The model is tuned until the marginal alert costs roughly what it saves. An anti-money laundering model has no such arithmetic available. Its output is a suspicion handed to the state, and the value of a report is not knowable by the institution that files it. What it optimises instead is coverage and defensibility: can we show that activity of this kind would have been seen.

So the cost of being wrong points in opposite directions. A fraud false positive is paid immediately, by a customer, in front of you. An AML false positive is paid by an analyst working an alert that goes nowhere — expensive in aggregate, nearly invisible case by case. A fraud false negative is a loss you can count by Friday. An AML false negative may never be discovered at all, and where it is, usually by a regulator.

What follows from that. Put both into one scoring engine behind one threshold and the fraud arithmetic wins, because it is the only one that can be stated in money. The AML side becomes a handful of rules bolted to a loss model, and the first thing to be tuned away is precisely the alert that costs something to review and saves nothing.

What to share instead: facts, not verdicts. The device seen across eleven customers, the beneficiary named in last month's scam reports, the account opened nine days before it received its first payment — these are observations, useful to both functions. Share the observation; let each side weigh it against its own objective. This is the practical end of the Wolfsberg Group's argument about monitoring effectiveness: the test of a monitoring arrangement is whether it produces output that somebody can act on, not whether every conceivable rule has been switched on.

One warning about the cleverer approaches. The FATF's work on new technologies is careful on this point — better tools pay off where the governance underneath them is already sound, and not otherwise. A shared model nobody can explain, trained on two data sets joined without a recorded basis, is not convergence. It is both functions' problems in one system, with fewer people able to see them.

Lesson 6 of 6

What goes wrong, and what to measure

The failures, as patterns:

  • The account closed too early. Fraud closes it, the money is gone, the network behind it is never mapped and the report is thin. The commonest and most expensive failure.
  • Fraud pays and moves on. Where a reimbursement regime makes paying the victim the cheapest resolution, the incentive is to settle and close. The suspicion is never reported, so the receiving institution and the financial intelligence unit learn nothing, and the mule account carries on.
  • AML ignores fraud alerts. "Fraud is not a predicate for us" — which is wrong almost everywhere, and produces an AML programme blind to one of the largest sources of proceeds it will ever see.
  • First-party fraud booked as credit loss. No report, no typology, no intelligence; just a write-off.
  • Insider matters handled only as employment issues. Resignation accepted, no report, the person joins a competitor.
  • Tipping off through routine fraud contact. Nobody intended it and the damage is the same.
  • Convergence as a reorganisation. New structure, same two queues, worse expertise, and a project declared complete.

What to measure. Not "number of joint meetings". Things that would be uncomfortable to report:

  • How many fraud cases produced a suspicion report, and how many did not and why. The single most revealing number in this subject.
  • How many accounts were closed by one function while the other had an open matter. Should be near zero, and will not be at first.
  • How many AML reports cited a fraud signal — a device shared across customers, a known scam beneficiary — that only the fraud team held.
  • Time from fraud detection to AML notification, in minutes or hours rather than days.
  • How many mule accounts were found by fraud, by AML, and by both. The overlap tells you whether you are duplicating or covering different ground.
  • How many first-party fraud and insider matters were assessed for reporting, rather than closed as loss or as personnel.

The judgement to hold on to. Fraud and anti-money laundering are not two risks that happen to share a customer base. They are one chain seen from two ends: a crime that takes money from somebody, and the movement of that money afterwards. The institution is usually the only party that can see both ends — and whether it does is an organisational choice, not a technical one.

Knowledge check

Twenty questions

Answer all twenty questions, then check your answers. You need 18 out of 20 to pass and receive a certificate. If you score less, you will see which answers were right and wrong, and then go through the course again before you retake the check. Your answers, progress and times are kept only in this browser.

Standards and frameworks covered

The published frameworks this course teaches toward. Saolix is not affiliated with, endorsed by or accredited by any of these bodies; naming a standard is a citation, not a claim about its author.

Sources

The official documents this course relies on. Laws and guidance change, so check the current version.

  1. Illicit Financial Flows from Cyber-enabled Fraud (November 2023) · FATF, Egmont Group and INTERPOL
  2. The FATF Recommendations (Recommendations 1, 10, 20 and 21) · Financial Action Task Force
  3. Statement on Effective Monitoring for Suspicious Activity, Part I: Moving Beyond Automated Transaction Monitoring · The Wolfsberg Group
  4. Statement on Effective Monitoring for Suspicious Activity, Part II: Transitioning to Innovation · The Wolfsberg Group
  5. Opportunities and Challenges of New Technologies for AML/CFT · Financial Action Task Force
  6. 2025 Report on Payment Fraud (December 2025) · European Banking Authority and European Central Bank
  7. Revised Recommendation 16 on payment transparency (June 2025) · Financial Action Task Force
  8. GovS 013: Counter Fraud (Government Functional Standard) · Cabinet Office (UK)
  9. FATF Guidance on the Risk-Based Approach and De-risking · Financial Action Task Force