How I Caught a Massive Money Laundering Ring in Bangladesh Using Graph Analysis

- May 15, 2026

Sitting at my desk, sipping coffee, I stared at the screen in front of me - 10,000 suspicious transactions in one day, all passing the BDT 100,000 threshold. I was working as an AML compliance analyst at a leading fintech in Bangladesh, and this was not what I wanted to see on a Monday morning.

The transactions were spread across various mobile financial services (MFS) like bKash, Nagad, and Rocket. We had to report these to the Bangladesh Financial Intelligence Unit (BFIU) as soon as possible. But how do we connect the dots and figure out if these are part of a larger money laundering network?

The Hidden Problem

Standard approaches to detecting money laundering often rely on simple threshold monitoring and rule-based systems. But these can generate a huge number of false positives, overwhelming the compliance team. In Bangladesh, the MFS threshold monitoring is a critical component of AML regulations, but it's not enough on its own.

That's why we need to use graph analysis to identify complex patterns and relationships between transactions. But this is not a straightforward task, especially when dealing with large datasets and complex networks.

Technical Breakdown & Logic Flow

So, how do we build a graph analysis system for money laundering detection? First, we need to create a data model that represents the transactions, customers, and their relationships. We can use a graph database like Neo4j to store this data.

Next, we need to define the rules for creating relationships between nodes in the graph. For example, if two customers have sent money to each other, we can create an edge between their nodes. We can also use other information like IP addresses, device IDs, and location data to create additional relationships.

Once we have the graph data model in place, we can start analyzing the network structure to identify potential money laundering patterns. This can include techniques like community detection, centrality analysis, and anomaly detection.

import networkx as nx
import pandas as pd

# Load transaction data
transactions = pd.read_csv('transactions.csv')

# Create a graph
G = nx.Graph()

# Add nodes and edges to the graph
for index, row in transactions.iterrows():
    G.add_node(row['customer_id'])
    if row['recipient_id'] in G.nodes():
        G.add_edge(row['customer_id'], row['recipient_id'])

# Analyze the graph structure
communities = nx.algorithms.community.girvan_newman(G)
centrality = nx.algorithms.centrality.degree_centrality(G)

As you can see, the code is quite straightforward, but the logic behind it is complex. We're using the NetworkX library to create and analyze the graph, and the Pandas library to load and manipulate the transaction data.

Local Application

So, how does this fit into the BFIU guidelines and MFS realities in Bangladesh? Well, the BFIU requires all MFS providers to monitor transactions above the BDT 100,000 threshold and report suspicious activity. By using graph analysis, we can identify complex patterns and relationships that may not be apparent through simple threshold monitoring.

BFIU guidelines require MFS providers to report suspicious transactions to the authorities as soon as possible.

In addition, the MFS providers in Bangladesh have to deal with a large volume of transactions, and manual review of these transactions is not feasible. By automated graph analysis, we can quickly identify potential money laundering patterns and prioritize the review of suspicious transactions.

Common Pitfalls & Edge Cases

One of the common pitfalls of graph analysis is the risk of false positives. If we're not careful, we can end up identifying legitimate transactions as suspicious. To avoid this, we need to carefully tune our algorithms and thresholds to minimize false positives.

Another edge case is the risk of overfitting. If we train our models on a specific dataset, they may not generalize well to new, unseen data. To avoid this, we need to use techniques like cross-validation and walk-forward optimization to ensure that our models are robust and generalize well.

Counterintuitive Insight

One of the counterintuitive insights I've gained from working on graph analysis for money laundering detection is that sometimes, the most suspicious transactions are the ones that look perfectly legitimate. For example, a transaction that is just below the BDT 100,000 threshold may be more suspicious than one that is above it, because it may indicate an attempt to avoid detection.

Conclusion & CTA

In conclusion, graph analysis is a powerful tool for detecting money laundering patterns in Bangladesh. By using graph databases and algorithms, we can identify complex relationships and patterns that may not be apparent through simple threshold monitoring.

So, what's the weirdest transaction pattern you've seen? Drop a comment below and let's discuss. Have you used graph analysis for AML detection? What were some of the challenges you faced, and how did you overcome them? Check out other resources on aitipseveryday.com for more information on graph analysis and AML detection.

Read about the latest developments in graph databases and AML regulations.
Check out our webinar on graph analysis for AML detection.
Download our whitepaper on the use of graph analysis in AML compliance.

Search

AML Data with Python