Building AI-ready data: Vanguard’s Virtual Analyst journey

This article describes how Vanguard built Virtual Analyst, a conversational AI solution for financial data analysis, by establishing AI-ready data infrastructure using AWS services.

• Vanguard faced slow data access requiring SQL expertise and multi-day response times from data teams
• Project revealed AI success depends on data architecture, not just foundation models
• Cross-functional teams (data engineers, analysts, compliance, security) collaborated on solution
• Eight guiding principles: clear data products, governance/security, metadata catalog, semantic layer, ground truth examples, data quality checks, change control, continuous evaluation
• Used Amazon Bedrock, Redshift, Glue, SageMaker, ECS, DynamoDB, S3 for implementation
• Results: reduced query time from days to minutes, enabled non-SQL users, high accuracy, decreased data team workload
• Future plans include knowledge graphs and Retrieval-Augmented Generation enhancements

Vanguard's approach demonstrates that enterprise AI success requires disciplined data foundations, cross-functional collaboration, and treating data as a strategic asset.