Procurement and supply chain teams believe they have diversified sourcing because their Enterprise Resource Planning (ERP) system shows multiple Tier-1 suppliers across different countries. But that data is incomplete: visibility ends at the first tier. When a disruption occurs at Tier-3, organizations are blindsided months later by sudden shortages, leaving no time to qualify alternatives.

This solution fuses internal ERP data with external trade intelligence into a heterogeneous knowledge graph, then trains a GraphSAGE model to infer common Tier-2+ supplier relationships, propagate risk scores through the network, and surface concentration risks in an interactive Streamlit dashboard. A Cortex Agent with a semantic model enables natural language risk queries through Snowflake Intelligence, all built and deployed entirely on Snowflake.

Traditional ERP visibility ends at Tier-1. Risks fester unseen in deeper layers of the supply network, creating costly blind spots for procurement teams.

This solution transforms supply chain management from reactive response to proactive resilience by fusing internal ERP data with external trade intelligence (customs-based shipment records, trade flow data, and geopolitical risk signals) into a knowledge graph that reveals what your ERP cannot see. The demo uses synthetic data generated in Snowflake; in production, this external intelligence would come from Snowflake Marketplace providers.

The platform constructs a heterogeneous knowledge graph with suppliers, parts, and regions as nodes and transactions and trade flows as edges. A GraphSAGE model trained on trade patterns infers likely Tier-2+ supplier relationships, then propagates risk scores through the network so that a shock at Tier-3 surfaces its impact on Tier-1 and final products.

1. Ingest. Integrate ERP data (vendors, materials, purchase orders, Bills of Materials) and external trade intelligence into Snowflake tables. The demo generates all six tables synthetically via a stored procedure.

2. Build the Graph. Construct a heterogeneous knowledge graph with suppliers, parts, and regions as nodes; transactions and trade flows as edges.

3. Infer Common Links. Train a GraphSAGE model on trade patterns to infer likely common Tier-2+ supplier relationships with probability scores.

4. Propagate Risk. Calculate risk scores that flow through the network so that a shock at Tier-3 surfaces its impact on Tier-1 and final products.

5. Visualize and Act. Explore the supply network graph, analyze concentration points, and prioritize mitigation actions in an interactive 8-page Streamlit dashboard.

6. Query with Natural Language. A Cortex Agent backed by a semantic model lets users ask risk questions in plain English through Snowflake Intelligence or the in-app Command Center.

The solution provides two notebook options: a GPU notebook using PyTorch Geometric on container runtime for full GNN training, and a NetworkX notebook running on standard warehouse compute (no compute pool or External Access Integration required) for trial accounts and environments where GPU is unavailable. Both produce compatible output for the dashboard.

Traditional analytics show a diversified supply base. Graph intelligence reveals the convergence: multiple seemingly independent Tier-1 suppliers may all depend on the same common Tier-2 refinery.

Before: "We're safe. We source from three different vendors in three countries."

After: "All three vendors rely on one Tier-2 supplier in a high-risk region. We need to qualify alternatives now."

Unified Data Foundation: Internal ERP data and external trade intelligence join seamlessly in a governed platform, with no data movement or pipeline complexity.

Performance That Scales: GPU-enabled notebooks train graph neural networks on millions of trade records without infrastructure friction.

Collaboration Without Compromise: Share risk insights with sourcing partners and internal teams while maintaining data governance and access controls.

Built-in AI/ML and Apps: From PyTorch Geometric models to interactive Streamlit dashboards, build and deploy intelligence closer to where decisions happen.

The solution fuses internal ERP data with external trade intelligence into a knowledge graph that reveals what your ERP cannot see. The following data represents the demo environment and is representative of production supply chain data.

The demo uses synthetic data to demonstrate the architecture end-to-end. Moving to production requires replacing synthetic tables with high-quality external data. These Snowflake Marketplace providers deliver the trade intelligence and entity data needed to operationalize N-tier visibility, with no data pipelines to build or contracts to negotiate outside your Snowflake account.

S&P Global Market Intelligence — Panjiva Supply Chain Intelligence: The gold standard for inferring common Tier-2+ relationships. Customs-based bill-of-lading shipment records with shipper/consignee entities, HS codes, origin/destination ports, volumes, and values. When your Tier-1 supplier appears as a consignee, Panjiva reveals who shipped to them, exposing your likely Tier-2 suppliers. Panjiva Supply Chain Intelligence | All S&P Global Listings

Oxford Economics — TradePrism: Forward-looking trade forecasts and scenario modeling. Global trade forecasts at HS4 level across ~170 economies, including tariff scenarios, sanction impacts, and trade rerouting projections. Combine historical shipment data with TradePrism forecasts to answer: "If sanctions expand to Region X, which of my supply chains are exposed?" TradePrism Full Dataset

FactSet — Supply Chain Linkages & Entity Data: Entity resolution and corporate relationship mapping. Map shipper/consignee names from trade data to actual vendor entities in your ERP and build supplier/geo exposure roll-ups at the corporate parent level. FactSet on Snowflake Marketplace

Resilinc — EventWatch AI: Real-time disruption monitoring and supplier risk signals. AI-powered monitoring of global events (natural disasters, geopolitical incidents, factory fires, labor actions) mapped to supplier locations and impact zones. Resilinc EventWatch AI

Exiger — Supply Chain Risk Management: End-to-end supply chain risk visibility and compliance monitoring. Exiger maintains one of the largest databases for multi-tier supplier mapping, providing validated Tier-2+ relationships, sanctions screening, and regulatory compliance signals.

Trademo — Global Trade Intelligence: Comprehensive import/export shipment data with buyer-supplier linkages across global trade corridors. Useful for validating GNN-inferred relationships against actual trade flow evidence.

Vesper — Commodity Intelligence: Real-time commodity market intelligence and supply-demand analytics. Enriches risk models with upstream commodity exposure and price volatility signals.

Why Marketplace Data Matters:

The 8-page Streamlit application guides users from executive summary to prioritized actions.

Ready to uncover common supplier dependencies and concentration risks in your supply chain? This guide includes everything you need to get up and running quickly.

GitHub Repository →

The repository contains the SQL setup script, two notebook options (GPU with PyTorch Geometric on container runtime, or warehouse-based with NetworkX), an 8-page Streamlit dashboard, a Cortex Agent with semantic model, and a teardown script for deploying the full solution.