PometryDocs
GitHub

Cookbook: Detecting Financial Fraud Rings

A complete walkthrough from raw transactions to fraud ring detection.

This cookbook demonstrates how to detect coordinated fraud - accounts created together that immediately transact with each other - using temporal graph analysis. We'll use synthetic transaction data that mirrors real-world financial crime patterns.

The Challenge

Financial fraud often involves coordinated account creation: bad actors create multiple accounts within a short time window, then use them to move money in layered transactions. Traditional rule-based systems struggle because:

  1. Individual transactions look legitimate
  2. Static graph analysis misses the temporal coordination
  3. The patterns only emerge when you consider time + structure together

What we'll detect:

  • Accounts created within the same hour
  • That immediately transact with each other
  • With suspicious flow patterns (rapid layering, circular flows)

The Data Model

Loading graph...

Step 1: Load the Dataset

We'll use a synthetic dataset representing 30 days of banking transactions. In production, this would come from your data warehouse.

Output:

Step 2: Build the Temporal Graph

Convert the DataFrame into a Raphtory temporal graph. Each edge carries the transaction timestamp - this is what enables temporal analysis.

Output:

Step 3: Identify Coordinated Account Creation

Fraud rings often involve accounts created within a tight time window. We look for nodes that "appear" in the graph at similar times.

Output:

Why time windows matter: Legitimate accounts are created throughout the day. Fraud rings create accounts in bursts - often within the same session or scripted process.

Step 4: Analyze Internal Connectivity

Coordination alone isn't fraud. We need to check if these accounts immediately transact with each other - a key indicator of layering.

Output:

Step 5: Trace Money Flow with Temporal Reachability

Once we identify a suspicious source, we trace where the money goes - respecting time order. Money can't flow backward.

Step 6: Generate Investigation Report

Compile findings into an actionable format for investigators.

Output:

Summary

This cookbook demonstrated a complete fraud ring detection pipeline:

StepWhat We Did
1. Load DataIngested transactions from pandas DataFrame
2. Build GraphCreated temporal graph preserving transaction times
3. Find CoordinationIdentified accounts created in tight time windows
4. Analyze ConnectivityMeasured internal transaction density and speed
5. Trace FlowFollowed money movement respecting temporal order
6. Generate ReportCompiled findings into investigation-ready format

Key temporal insights that static analysis misses:

  • Time-ordered flow: Money can't move backward in time
  • Creation coordination: Fraud rings create accounts in bursts
  • Rapid activation: Legitimate accounts don't transact immediately at high volume

Next Steps