Pre-Disbursement Fraud Detection for Multi-Tier Supply Chain Finance
Detecting fraud hidden across supplier networks before financing approval.
"Each invoice looked valid. Together, they formed a $47M fraud." ChainGuard reveals risks invisible to isolated invoice checks.
Supply Chain Finance (SCF) allows lenders to release funds based on supplier invoices across multiple tiers:
Tier-1 → Tier-2 → Tier-3 suppliers
A real-world fraud case demonstrated:
- 340 fabricated invoices worth ~$47M
- Each invoice passed traditional validation
- Multiple lenders financed the same economic activity
- Goods never existed
- Circular trading masked exposure
- Fraud became visible only after defaults occurred
Traditional systems validate documents. Fraud emerges from relationships.
ChainGuard is a pre-disbursement fraud intelligence layer designed for lenders.
It answers a single decision-critical question:
Should this invoice be trusted before funds are released?
The system correlates invoices, suppliers, transactions, and behavioral signals across the entire supply chain network and produces an explainable risk score.
ChainGuard is built on three core assumptions:
- Fraud manifests at the network level, not document level.
- Economic infeasibility appears before financial default.
- Data silos between lenders create systemic blind spots.
Therefore, detection must analyze relationships, timing, and feasibility simultaneously.
1️⃣ Supplier generates invoice
2️⃣ Lender finances invoice after validation
3️⃣ Same trade reappears across supply chain tiers
4️⃣ Multiple lenders unknowingly finance duplicates
5️⃣ Circular trade loops hide fraud exposure
6️⃣ Losses appear only after payment failure
Fraud becomes visible only when the entire network is analyzed together.
ChainGuard assumes adversaries may:
- Create syntactically valid invoices
- Coordinate across shell suppliers
- Exploit lender data isolation
- Inflate invoice velocity gradually
- Recycle invoices via circular trade loops
Detection therefore targets behavioral inconsistencies, not formatting anomalies.
| Fraud Type | Description |
|---|---|
| 🧾 Phantom Invoices | Fabricated trade activity |
| 🔁 Double Financing | Same invoice funded multiple times |
| 📈 Over-Invoicing | Value exceeds economic capacity |
| 🔄 Carousel Trades | Circular A → B → C → A transactions |
| 💧 Dilution Fraud | Manipulated post-collection reporting |
| 🌊 Cross-Tier Cascading | Exposure multiplied across tiers |
┌─────────────────────────────────────────────────────────────┐
│ DATA INGESTION LAYER │
│ Invoices │ Purchase Orders │ GRN │ ERP Feeds │ Payments │
└─────────────────────────┬───────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────┐
│ MULTI-LAYER FRAUD DETECTION ENGINE │
│ │
│ Layer 1: Invoice Fingerprinting & Deduplication │
│ Layer 2: Feasibility Scoring (Revenue vs Volume) │
│ Layer 3: Velocity & Sequencing Analysis │
│ Layer 4: Graph Analysis — Cycles & Cascades │
└─────────────────────────┬───────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────┐
│ RISK INTELLIGENCE LAYER │
│ Weighted Signal Aggregation → Risk Score (0–100) │
└─────────────────────────┬───────────────────────────────────┘
│
┌─────────────┼─────────────┐
▼ ▼ ▼
🟢 Low 🟡 Medium 🔴 High
Auto Approve Manual Review Block
Supply chains naturally form networks:
- Nodes → Companies
- Edges → Transactions
- Cycles → Potential fraud loops
Graph traversal enables detection of circular trading, abnormal relationship density, and disconnected economic flows.
Tabular validation cannot capture these structures.
| Step | Module | Purpose |
|---|---|---|
| 1 | Invoice Input | Financing request received |
| 2 | Fingerprinting | SHA-256 invoice identity check |
| 3 | Graph Analysis | Multi-tier relationship mapping |
| 4 | Feasibility Checks | Economic realism validation |
| 5 | Behavioral Analysis | Velocity & sequencing anomalies |
| 6 | Risk Scoring | Explainable fraud probability |
| 7 | Alert Engine | Pre-disbursement decision |
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
CHAINGUARD RISK REPORT
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Invoice ID : INV-2024-00847
Supplier : XYZ Trading Co.
Amount : ₹4,80,00,000
Risk Score : 82 / 100 🔴 HIGH
Flags:
✗ Duplicate invoice across lenders
✗ Invoice volume exceeds revenue by 40×
✗ Circular trade detected
Recommendation: BLOCK DISBURSEMENT
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
| Layer | Technology |
|---|---|
| Backend | Python, FastAPI |
| Graph Engine | NetworkX / Neo4j |
| Analysis | Statistical + Rule-Based Models |
| Frontend | Streamlit / React |
| Data | Synthetic SCF datasets |
| Fingerprinting | SHA-256 hashing |
Prototype performance evaluated using:
- Fraud detection recall across simulated cascades
- False positive rate per supplier tier
- Detection latency before disbursement
- Network anomaly precision
Goal: maximize early detection while minimizing operational friction.
- Synthetic datasets used for prototype testing
- Cross-lender data simulated
- Risk weights manually calibrated
- Real-time streaming under development
- 🛑 Prevent fraud before funds leave lenders
- 🔁 Reduce duplicate financing exposure
- 🕸️ Detect hidden network-level fraud
- 📊 Provide explainable risk decisions
- 📡 Scale across multi-tier ecosystems
- Real-time ERP integrations via REST APIs
- Kafka-based streaming ingestion
- Adaptive ML risk calibration
- Multi-bank intelligence sharing
- Regulatory audit modules
Supply chain fraud is not a niche problem — it costs lenders and economies billions every year. Most of it goes undetected until after the money is gone.
We are working towards a world where no lender finances a fraudulent invoice simply because they lacked visibility into the network around it.
ChainGuard is our step in that direction — a system that treats fraud as a network problem, not a document problem, and catches it before damage is done.
This is a hackathon prototype today. The problem it solves is real, urgent, and largely unsolved at scale.
ChainGuard — Fraud hides in connections, not documents.