Reporting: CTRs, SARs, and STRs

What are the report types?

Every AML regime runs on the same three instruments, whatever the local acronyms.

Threshold reports are automatic. In the US, any cash transaction over $10,000 produces a Currency Transaction Report; Canada’s Large Cash Transaction Report triggers at C$10,000, and Australia’s Threshold Transaction Report at A$10,000. Casinos file them too. There is no judgment involved and no stigma: a car dealership’s good Saturday generates them. The customer is aware; the report is just a record that this much cash moved.

Suspicion reports are the opposite in every way. A Suspicious Activity Report (US) or Suspicious Transaction Report (Canada) is filed only when something looks wrong: deposits engineered around thresholds, activity that doesn’t fit the customer, money moving with no economic point. Thresholds are low ($5,000 for US banks, $2,000 for money services businesses) or absent entirely, as in Canada and the UK. And the filing is secret. In most jurisdictions, telling the customer a report exists is a crime called tipping off.

Border declarations cover physical movement: more than US$10,000 into or out of the US, £10,000 for Great Britain, €10,000 for the EU. Carrying the cash is legal. Failing to declare it converts legal money into seizable evidence. The full country-by-country table lives on the thresholds page.

How does a suspicious report actually get written?

Inside a bank, a report usually starts as an alert: monitoring software flags an account for a pattern (velocity, near-threshold clustering, geography) or a teller escalates something seen at the counter. An analyst reviews the alert against the customer’s profile and history. Most alerts die there as false positives. The survivors become a written narrative: who, what, when, where, and why it looks like laundering, with the transactions attached.

That narrative matters more than outsiders assume. Investigators searching the database years later will find the report by name or account; whether it helps depends on whether the analyst wrote down what they actually saw. Regulators publish guidance on narrative quality for exactly this reason.

The report then goes to the national financial intelligence unit (FinCEN, FINTRAC, AUSTRAC, the UK’s UKFIU) and into a database queried by investigators across government.

Why do most reports go nowhere?

Because the arithmetic is brutal. The US alone receives millions of suspicious activity reports a year (about 4.7 million in fiscal 2024) plus tens of millions of currency reports. No agency reads that in real time, and no agency is meant to. The system is not an alarm; it is a memory.

The economics push volume up. A bank that under-files ends up like TD Bank (a US$3 billion resolution in 2024 built on years of monitoring failures), while a bank that over-files pays only in analyst hours. Defensive filing is the rational response, and the strong reports sink into the noise.

What makes the memory valuable is the query. When a name surfaces in a drug case, a fraud complaint, or a leak, investigators pull every report ever filed on it, and suddenly a decade of scattered suspicions becomes a map. Nearly every major case in the case files rests on records that reporting rules forced institutions to create and keep.

How reporting catches launderers

Directly, three ways. Aggregation rules defeat structuring by combining transactions across branches and days, so the pattern surfaces no matter how the deposits were sliced. Cross-institution matching at the FIU connects mule networks that no single bank can see. And border declarations give customs a lawful hook to search, seize, and prosecute cash movements that would otherwise be invisible.

Indirectly, and more importantly, reporting changes the launderer’s problem. Every technique on this site is, at bottom, an attempt to move value without generating an honest report. That constraint is what makes laundering expensive, slow, and error-prone. The reports don’t have to catch everyone; they have to make not being caught cost something.

First published August 28, 2026 · Last updated August 28, 2026