Sankey diagram

A Sankey diagram is a specific type of flow diagram where the width of the arrows is proportional to the flow quantity (e.g., data volume, funds, energy), visually representing the distribution and movement of quantities between nodes in a system. Originally used by engineer Matthew Sankey in 1898 to analyze steam engine efficiency, it is now a powerful tool in risk management. While not defined by a specific standard, its application is highly relevant to regulatory compliance. For instance, it can be used to map data processing activities to meet the record-keeping requirements of GDPR Article 30. In AI governance, it visualizes data lineage from collection to model inference, helping to trace sources of bias and supporting the principles of traceability and explainability in frameworks like the NIST AI Risk Management Framework (AI RMF) and the transparency requirements of ISO/IEC 42001. Its key distinction from a standard flowchart is its ability to quantify flows, revealing bottlenecks, losses, or major impact pathways.

In enterprise AI risk management, a Sankey diagram offers a powerful visualization method. The implementation involves these steps: 1. **Define Scope**: Identify the AI process to be analyzed, such as the data flow for an AI credit scoring model or the risk propagation path in a facial recognition system. Clearly define the start, end, and all intermediate nodes. 2. **Collect & Quantify Data**: Gather flow data between nodes. For bias analysis, this could be the number of data records from different demographic groups. For risk analysis, it could be quantified risk values propagating from various sources to business impact points. 3. **Visualize & Interpret**: Use tools like Python (e.g., Plotly), Tableau, or Power BI to create the diagram. Analyze the widths and paths of the flows to identify data bottlenecks, stages where bias is amplified, or critical risk pathways. **Real-world Example**: A tech firm used a Sankey diagram to analyze applicant data flow in its AI recruitment tool. The diagram clearly showed that resumes from specific universities were disproportionately rejected at an early stage, revealing a potential educational bias. After adjusting the algorithm, the model's fairness metrics improved by 25%, ensuring compliance and reducing legal risks.

Taiwanese enterprises often face three main challenges when implementing Sankey diagrams for AI governance: 1. **Data Silos and Poor Quality**: Data is often fragmented across legacy systems in different departments with inconsistent formats, making it difficult to collect accurate end-to-end flow data. This can result in a distorted or misleading Sankey diagram. 2. **Lack of Hybrid Talent**: Effective use of Sankey diagrams requires a combination of business process knowledge, data analysis skills, and proficiency with visualization tools. Many companies lack experts who can interpret these diagrams to uncover AI ethics risks or compliance gaps. 3. **Poorly Documented Processes**: The data processing workflows for many AI applications, especially those spanning multiple departments, are often not clearly documented. This makes it extremely difficult to accurately define the nodes and flow paths for the diagram. **Solutions**: * **Establish a Data Governance Committee**: Form a cross-functional team to promote data standardization and centralized management (e.g., a data middle platform). Priority action: Conduct data mapping for high-risk AI applications. Expected timeline: 6 months. * **Adopt BI Tools and Empower Staff**: Implement user-friendly BI tools and conduct workshops to train employees in data visualization and risk interpretation. * **Engage External Consultants**: Partner with experts like Winners Consulting to quickly perform business process analysis (BPA) and create clear data maps, providing a solid foundation for creating accurate Sankey diagrams.

Winners Consulting specializes in AI governance and data risk management for Taiwan enterprises, with extensive hands-on experience in using Sankey diagrams. We help companies build management systems compliant with international standards like NIST AI RMF and ISO/IEC 42001 within 90 days. Free consultation: https://winners.com.tw/contact