Structuring: the oldest trick, still the hardest to prove

Each month, we spotlight a real-world AI task in Unit21. This month: structuring, the practice of splitting cash into amounts that stay under the $10,000 reporting line.

It is one of the oldest tricks in financial crime, and one of the hardest to prove. A $9,400 deposit doesn't look wrong. Neither does $9,100, $9,800 or $9,300. Each one sits under the threshold, and the pattern is the only thing that gives it away. So we ran structuring through Unit21 twice: once to catch it, and once to prove it.

‍

‍

What it does

Structuring shows up across cash, ACH and crypto, and no single transaction raises a flag. Catching it means describing the pattern precisely enough to detect it. Proving it means turning a long run of transfers into something an examiner will accept. This edition covers both halves.

‍

Part 1: Rule Writer Agent. Plain English in, deployable rule out. We described the behavior the way an analyst would say it out loud: three or more approved transactions, each between $8,000 and $9,999, adding up to more than $10,000 within a rolling three-day window, individuals only. Rule Writer Agent built the rule. From there it follows the same path as any other rule: edit it, validate it, shadow test it, then go live.

‍

Part 2: AI Investigation Agent. On a flagged alert, the AI review does the legwork before an analyst opens the case. It spotted strong structuring indicators: 174 transfers, $53,871 in total, to a single recipient over 25 days. Four larger transfers were hidden among 169 small ones, and 17 of the transfers happened in a single day. The agent's recommendation: human review.

‍

The agent automatically reviews:

• Transaction amounts relative to the $10,000 reporting threshold
• Timing and frequency across a rolling window, including bursts within a single day
• Recipient concentration, such as many transfers to the same counterparty
• Prior alerts and historical rule patterns for the same entity
• Watchlist analysis tied to the alert

‍

What the agents output:

• Rule Writer Agent: a complete trigger condition generated from plain English, ready to edit, validate and shadow test before anything goes live
• AI Investigation Agent: findings traced to the underlying transactions, so every claim can be checked
• A clear recommendation, in this case human review requested
• An editable narrative drafted for the analyst to refine
• A full audit trail of every step the agent took

‍

Why this matters

Structuring is hard to prove because the evidence is spread across dozens of transactions that each look ordinary. The work is in the assembly: finding the pattern, counting the transfers, tying them to one recipient, and writing it up so it holds under examination. That assembly is where analyst time goes.

‍

Here, detection started with a few sentences of plain English, and the investigation ended with an editable narrative and a complete audit trail. The agent recommends. The analyst decides.

Describe it. Detect it. Investigate it.

‍

See it live: the structuring masterclass

We're walking through structuring live in our AI in Fincrime Masterclass on Tuesday, October 13 at 10:00 AM PT. You'll see where structuring shows up today, how Unit21 works, the patterns and dollar thresholds examiners expect programs to catch, and a live look at Rule Writer Agent and SAR Agent turning a flagged pattern into a filing. No prior Unit21 experience needed.

‍

Register for the structuring masterclass →

The series continues with adverse media and sanctions screening on October 27 and mule networks on November 17. Sign up for the full series here.

‍