Jefferies deploys agentic AI platform JDI to accelerate equity research from days to minutes

Jefferies' equity research operation spans coverage of roughly 3,500 companies across global sectors—a breadth that had grown increasingly difficult to service through manual workflows. Analysts relied on custom SQL queries, ad-hoc data requests routed through support teams, and labor-intensive cross-referencing of fundamentals, alternative datasets, and macroeconomic indicators like FRED and BLS releases. As data volumes exploded, these bottlenecks compressed the time available for the high-conviction analysis that differentiates elite research. Synthesizing signals across disparate sources at that scale was no longer sustainable, limiting both analyst throughput and the range of signals any single analyst could realistically consider.

Jefferies partnered with Databricks to build Jefferies Data Intelligence (JDI), a conversational analytics platform combining Databricks AI/BI Genie with a custom LangGraph-based multi-agent architecture. The agentic system decomposes natural-language research questions through four specialized agents: a validation agent that confirms tool and API availability; a planning agent that breaks complex queries into parallelizable tasks; concurrent execution agents pulling from structured fundamentals, alternative data (web traffic, foot traffic, social engagement), and real-time macroeconomic sources; and a synthesis agent that assembles auditable responses with chain-of-thought transparency. Governed by Unity Catalog and built atop Jefferies' existing seven-year Databricks data ingestion pipelines, JDI avoided infrastructure duplication and reached production deployment rapidly—rolling out to over 250 U.S. analysts within weeks of launch.

JDI delivered measurable acceleration immediately after rollout:

• 250+ analysts onboarded across U.S. equity research within weeks of launch
• Research synthesis time cut from days or weeks to minutes for complex, multi-source queries
• Hundreds of queries fielded and thousands of insights and charts generated in early weeks of deployment
• Expansion target of ~550 analysts across EMEA and APAC underway
• Data source coverage growing from 10–12 core datasets toward 30–40 or more

Beyond speed, analysts reported exposure to signals they would not have surfaced manually, strengthening thesis conviction through cross-corroboration of independent datasets rather than relying on a single source.

• Build on existing data infrastructure: Layering agentic AI atop Jefferies' seven-year Databricks pipelines eliminated duplication and dramatically shortened deployment timelines.
• Decompose, don't monolith: A multi-agent architecture—validation, planning, parallel execution, synthesis—outperforms single-model approaches for complex, multi-source research questions.
• Corroboration beats consolidation: Routing the same question through multiple independent datasets strengthens analyst conviction in ways that simple data aggregation cannot replicate.
• Natural language lowers the adoption barrier: Removing SQL expertise as a prerequisite enabled rapid onboarding of 250+ analysts without extended training cycles.
• Design for global scale from day one: Architecting JDI to be model-agnostic and pipeline-reusable made EMEA/APAC expansion a configuration decision, not a rebuild.