Kensho's Grounding Framework Unifies Financial AI Data

Financial professionals spend countless hours hunting through disconnected databases to find verified information for earnings analysis, market research, and compliance checks. This fragmented data landscape creates inefficiencies and risks in an industry where accuracy and trust are paramount. Kensho, SP Global's AI innovation team, has developed a solution that could transform how financial institutions interact with their most valuable asset: structured data.

Kensho created Grounding, a multi-agent framework that serves as a centralized access layer for SP Global's vast financial datasets. The system allows users to submit natural language queries and receive citation-backed responses from verified sources, eliminating the need to navigate complex database schemas or learn specialized query languages. This approach ensures every insight is derived directly from trusted data while maintaining the transparency and traceability required in financial services.

Ology involves a sophisticated router architecture built using LangGraph tools that intelligently directs queries to specialized Data Retrieval Agents (DRAs). These DRAs are owned by different data teams across equity research, fixed income, macroeconomics, and other financial domains. Rather than embedding natural language parsing logic into individual agents, the router breaks down queries into DRA-specific sub-queries, then aggregates responses into coherent insights while maintaining accuracy and context.

A critical innovation is the custom DRA protocol that establishes consistent data formats for both structured and unstructured data returns. This protocol solves the communication interface problems that typically plague distributed AI systems, ensuring reliable agent interactions and accelerating collaboration across the multi-agent ecosystem. The protocol's standardization has enabled Kensho to rapidly deploy multiple specialized financial AI products while maintaining data consistency.

Show significant practical applications already in production. Kensho has deployed an equity research assistant that helps analysts compare sector performance and an ESG compliance agent that tracks sustainability metrics, both built atop the same Grounding framework. Building agents on this consistent system accelerates time-to-market, giving new applications immediate access to SP Global's full data breadth without rebuilding data pipelines for each product.

The team implemented rigorous evaluation s to ensure reliability in the high-stakes financial environment. Their multi-stage evaluation suite executes test queries through the LangGraph-based RouterGraph, measuring accuracy against selected agents with focus on exact-match (correct agents and expected responses) and tool-calling metrics. This comprehensive testing approach allows the team to identify gaps and maintain the high trust requirements of financial data retrieval.

Kensho's experience yielded three key insights for organizations developing complex multi-agent architectures. First, comprehensive observability with deliberate metadata requirements is essential for maintaining visibility into agent behavior at scale. Second, multi-faceted evaluation matrices must assess routing decisions, data quality, and answer completeness at each retrieval step. Third, continuous analysis of user and agent interaction patterns enables iterative protocol refinement that improves system efficiency while maintaining financial-grade reliability.

The Grounding framework represents a significant step toward solving one of AI's most persistent s in regulated industries: maintaining data integrity while enabling natural language access. By creating a separation of concerns between data routing and retrieval layers, Kensho has developed a model that could influence how other data-intensive industries approach AI integration while meeting stringent compliance requirements.