M Science

M Science has rebuilt two decades of analyst-curated data for AI and quant use. Its Unified Data Model, launched in June 2026, puts more than 1,440 KPIs on over 1,400 companies into one star schema with change-data capture, so clients can test history point-in-time; it is delivered by Snowflake Share, Databricks Delta Sharing, S3, API and an MCP server for ChatGPT and Claude. Maddie, a multi-agent copilot launched in February 2026, adds alerts and anomaly detection on the same data, and analysts still publish company research two to three times a quarter.

Very little of its enterprise posture is public. We did not locate SOC 2 or ISO 27001, SSO or audit-log documentation, an AI-use or no-training statement, or a description of how sources are screened for consent and MNPI. Panel sizes are not disclosed, coverage figures vary between sources, no predictive-value study or quantified client case was located, and pricing is not published. Jefferies ownership gives continuity; a trade-secrets dispute with YipitData was settled privately in 2025.

We initiate at 6.20, Emerging, with an Improving outlook. Workflow Fit, at 7.75, is the highest in the segment; Enterprise Readiness (5.00) and Commercial Value (4.25) hold the score down. Published assurance and AI terms would move it most.

Strengths

• Point-in-time model. Change-data capture in the Unified Data Model supports point-in-time analysis, historical comparison and back-testing.
• Quant delivery. Snowflake Share, Databricks Delta Sharing, S3, API and more than 600 programmatic endpoints.
• AI-assistant delivery. An MCP server that carries analyst context into ChatGPT, Claude and in-house systems.
• KPI depth. More than 1,440 KPIs on over 1,400 companies across about 40 subsectors (company).
• Analyst research. Written company research and direct analyst engagement alongside the data.
• Continuity. Wholly owned by Jefferies, with research history back to 2002.

Cautions

• Assurance. We did not locate SOC 2, ISO 27001, SSO or audit-log documentation.
• AI terms. No AI-use, model-provider or no-training statement was located; buyers should request them before using the MCP server.
• Provenance. Source consent, MNPI and personal-data screening are not publicly documented.
• Panel disclosure. Panel sizes are not disclosed, and coverage figures differ between sources (250+ public companies in February 2026; 1,400+ companies today).
• Evidence. No published predictive-value study or quantified client case was located.
• Pricing. Not published; no trial located.
• Litigation history. A trade-secrets dispute with YipitData over hired staff was settled privately in 2025.