Cash structuring · one worked example

Three deposits.
None over $10,000.
All in one day.

Each deposit is under the reporting threshold, so each one passes on its own. Together they are a reporting obligation. The question a regulator asks two years later is not whether you had a rule. It is which rule was running that day, and whether it ran.

Move the money. Watch the decision change.

Three cash deposits from one customer, across one day. Drag any of them.

$3,800
$3,200
$2,600
Or try a real pattern
ALLOW

Aggregate under threshold.

Decision
ALLOW
Rule in force
AUSTRAC-TTR / 10,000 AUD / 24h
Rule version
v2026.03
Recorded

That last block is the point. It is what you hand a regulator to show which rule was in force when the decision was made.

Synthetic figures, running in this browser. The arithmetic is the same as the open-source gate: integer cents, a rolling 24-hour window, and a single deposit at or over the threshold reported in its own right rather than counted as a structuring catch.

The number nobody publishes

Catching structuring is the easy half.

356 of 356 structuring days caught
57% of legitimate cash businesses held for review
26,521 synthetic customer-days measured

Summing deposits and comparing to a threshold is arithmetic. Every transaction monitoring product has done it since the 1990s, and a rule that holds everything scores a perfect catch rate. The number that decides whether a rule can go anywhere near production is the second one: the market traders, restaurants and car washes whose ordinary Tuesday looks exactly like structuring.

That rule is not deployable as written, and that is the finding. Bringing the false-hold rate down without losing the catch rate is the actual work, and it can only be done against a real institution's transaction mix. That is what a pilot is for.

What changes with AI in the loop

The rule is not the new problem. Proving which rule ran is.

When a person applied the policy, the file showed their reasoning. When a model drafts the assessment, the same question is harder to answer: which policy version did it use, was it the one in force that day, and would it give the same answer if asked again.

EcoKure sits beside the existing process and answers those three questions for every decision. It does not replace transaction monitoring and makes no suspicion determination.

What a pilot looks like

One workflow, in shadow mode, for a quarter.

  1. Your risk function owns the threshold, the window and the policy citations.
  2. It runs beside production, changing nothing.
  3. You get the decisions, the exceptions and the false-hold rate on your own mix.
  4. Then a go or no-go against criteria agreed before the start.

What this is not. It is not an AML system. It makes no suspicion determination, files nothing, and does not replace transaction monitoring or a reporting obligation. The threshold and window shown here are one configured rule for demonstration, not AUSTRAC guidance. Every figure on this page comes from synthetic data; catch and false-hold rates depend entirely on a real institution's transaction mix. EcoKure holds no AUSTRAC or APRA approval and this page is not compliance advice.

Next · See it work Eco Control Tower