What is Trade-Based Money Laundering?

Trade-Based Money Laundering (TBML) exploits legitimate international trade to disguise illicit funds. Criminals manipulate prices, quantities, or quality specifications in trade transactions to move value across borders. TBML is estimated to account for $2-3 trillion annually—more than drug trafficking and human smuggling combined.

Common TBML Schemes

1. Over-Invoicing

Scheme: Importer pays exporter $500K for goods worth $50K. Exporter keeps $50K, launders $450K back to importer.

Result: $450K moved from Country A to Country B disguised as trade payment.

Common in: Capital flight from high-tax or restricted countries

2. Under-Invoicing

Scheme: Exporter ships $500K in goods but invoices only $50K. Receives $450K "under the table" to offshore account.

Result: Tax evasion and money laundering combined.

Common in: Import/export tax avoidance, VAT fraud

3. Phantom Shipping

Scheme: Invoice and payment for goods that are never actually shipped.

Detection: No corresponding bill of lading, shipping manifest, or customs entry.

4. Circular Trading

Scheme: Goods shipped in circle (A→B→C→A) with inflated prices at each step.

Result: Value artificially inflated 10x through circular route.

Why Traditional Detection Fails

Legacy TBML detection relies on manual review and simple rules:

Price Thresholds: Flag transactions where unit price deviates >50% from average
Country Pairs: Flag high-risk corridors (e.g., Venezuela → Panama)
HS Code Mismatch: Check if commodity codes align with descriptions

Problems with this approach:

95%+ false positive rate (legitimate price variations)
Easily evaded (criminals stay within threshold ranges)
No historical context (can't detect gradual price manipulation)
Data silos (payment systems don't talk to shipping systems)

AI-Powered TBML Detection

Our approach combines multiple data sources and ML techniques:

Data Integration

Trade Payments: SWIFT messages, wire transfers, letters of credit
Shipping Documents: Bills of lading, shipping manifests
Customs Data: Import/export declarations, HS codes
Market Prices: Commodity benchmarks, historical pricing
Entity Information: Company registries, beneficial owners

Pricing Analysis

Instead of simple thresholds, we build sophisticated pricing models:

Example: Electronics Import Pricing

Commodity: Laptop computers (HS code 8471.30)
Route: China → United States
Quantity: 1,000 units

Historical Price Distribution (90 days):
  Mean: $450/unit
  Std Dev: $50/unit
  p5: $380/unit, p95: $520/unit

Invoice Under Review:
  Declared price: $850/unit
  Z-score: (850-450)/50 = 8.0

Risk Assessment: HIGH
Reason: Price 8 standard deviations above mean
Likely over-invoicing to move capital

Network Analysis

Graph Neural Networks identify suspicious trading networks:

Circular Trading: Detect cycles in trade graph
Shell Companies: Identify entities with no physical presence
Related Parties: Link companies with common ownership
Hub Detection: Find intermediaries in layering schemes

Circular Trading Detection

Copper shipments form a cycle:

• Company A (Mexico) ships to Company B (Panama): $1M
• Company B ships to Company C (Colombia): $1.5M
• Company C ships to Company D (Brazil): $2.2M
• Company D ships to Company A (Mexico): $3.3M

GNN Detection: Identifies 4-node cycle with price inflation at each hop. Total value artificially inflated from $1M to $3.3M.

Anomaly Detection in Shipping

Match payment data to shipping records:

Phantom Shipping: Payment without corresponding bill of lading
Weight Discrepancies: Declared weight doesn't match shipping manifest
Routing Anomalies: Unnecessary detours through tax havens
Container Tracking: Verify physical movement matches documentation

Feature Engineering for TBML

Key features that power our TBML models:

Price Features

Z-Score: Deviations from commodity benchmark
Historical Comparison: Relative to this importer/exporter pair's history
Peer Comparison: Relative to similar companies trading same commodity
Market Alignment: Correlation with spot/futures prices

Relationship Features

Shared Ownership: Common beneficial owners or directors
Trading Frequency: How often these entities trade together
Payment Terms: Unusual terms (e.g., advance payment for untrusted partner)
Relationship Tenure: New vs established trading relationships

Geographic Features

High-Risk Corridors: Routes known for TBML (e.g., certain China-Latin America routes)
Tax Haven Involvement: Payments routed through offshore centers
Conflict Zones: Trade with sanctioned or high-risk jurisdictions

Machine Learning Pipeline

Data Ingestion: Collect trade payments, shipping docs, customs data
Entity Resolution: Link companies across different data sources
Price Normalization: Standardize units, currencies, quantities
Feature Extraction: Compute 300+ TBML-specific features
Ensemble Prediction:
- • Pricing Model: Anomaly detection on price deviations
- • Network Model: GNN on trading relationships
- • Document Model: NLP on invoice text, shipping docs
- • Rules Engine: Known typologies and red flags
Risk Scoring: Combine model outputs into unified risk score
Alert Generation: High-risk transactions create investigative cases

Real-World Case Studies

Case 1: Electronics Over-Invoicing

Scheme: Chinese exporter invoices smartphones at $1,200/unit. Market price: $250/unit.

Detection: Price anomaly model flags 380% over-invoicing. Investigation reveals capital flight scheme: Chinese company moving profits offshore.

Result: $45M illicit capital movement detected, SAR filed

Case 2: Circular Gold Trading

Scheme: Gold shipments cycle through 6 companies across 4 countries. Price inflates from $50M to $180M in one month.

Detection: GNN identifies circular trading pattern. All 6 companies trace to same beneficial owner (shell company network).

Result: $130M trade-based laundering scheme disrupted

Challenges in TBML Detection

Data Quality: Shipping docs often incomplete or inaccurate
Price Volatility: Legitimate prices vary widely for commodities
Cross-Border Data: Customs data not accessible across jurisdictions
False Positives: Luxury goods, custom orders have legitimately unusual prices
Definitional Ambiguity: Hard to distinguish TBML from aggressive tax planning

Performance Benchmarks

73%

Improvement in TBML detection vs traditional methods

68%

Reduction in false positives

$2.1B

Total suspicious trade value flagged by clients

24min

Average investigation time per alert

Regulatory Compliance

TBML detection must satisfy specific regulatory requirements:

FinCEN: Include trade finance in AML programs
FATF Recommendation 32: Monitor cash couriers and TBML
EU 5AMLD: Enhanced due diligence for trade transactions
US Customs: Report suspicious trade patterns

Future Enhancements

We're actively developing:

IoT Integration: Real-time container tracking via GPS/RFID
Satellite Imagery: Verify shipping against port activity
Cross-Institution Collaboration: Federated learning across banks
NLP on Invoices: Detect inconsistencies in invoice descriptions

Conclusion

Trade-Based Money Laundering represents one of the largest and least-detected forms of financial crime. Detecting it requires integrating diverse data sources, sophisticated pricing models, and network analysis—capabilities that traditional systems lack.

At nerous.ai, where our name reflects the Finnish spirit of ingenuity and genius, we've built AI systems that detect TBML schemes worth billions. By combining pricing analysis, network detection, and cross-border intelligence, we're helping financial institutions and customs agencies close a critical gap in global AML defenses.

Trade-Based Money Laundering Detection with AI