FactSet

FactSet has chosen to meet clients inside the AI tools they already use. Its MCP server, live since December 2025, gives Claude, ChatGPT, Gemini Enterprise and Microsoft Copilot governed, entitlement-checked access to estimates, fundamentals, transcripts and analytics; 650+ clients were on contracts or trials by August 2026. AI products supplied over 10% of fourth-quarter new subscription value. Pitch Creator and 200+ Excel templates make it strong for banking deliverables.

Its own assistant and its disclosure are behind. AI document search was still a phased beta in spring 2026, idea generation and monitoring are not yet agentic, no accuracy evaluation is published, and the 10-K says only that security is "guided by" ISO 27001 and SOC 2. Model providers inside FactSet products are not named, and pricing is not published.

We initiate at 6.57, Established, with an Improving outlook. Delivery surfaces and consensus data are among the strongest in our coverage; Commercial Value, at 5.17, is the weakest dimension. Published evaluation and security documentation, and an agentic monitoring layer, would move the score most. Investor Day on 10 November 2026 is the next catalyst.

Strengths

• Open distribution. Production MCP server across Claude, ChatGPT, Gemini Enterprise and Microsoft Copilot, with 650+ clients engaged.
• Consensus and estimates. FactSet Estimates with broker-level detail, available to agents through MCP.
• Banking deliverables. Pitch Creator, Slide Assistant and 200+ Excel templates with server-side refresh.
• Governed outputs. A semantic layer ties agent queries to audit-approved outputs, with sentence-level links in Portfolio Commentary.
• Commercial traction. AI annual subscription value added more than doubled in FY26 (company).
• Commitment. A Chief AI Officer, an AI Blueprint since 2023 and AI products supplying 10%+ of new subscription value.

Cautions

• Own assistant. AI document search was a phased beta into spring 2026; idea generation and monitoring are not agentic.
• Evaluation. No published accuracy evaluation; OpenAI reported FactSet's ChatGPT connector error rate at 6.45% after optimisation.
• Security documentation. The 10-K describes the programme as "guided by" ISO 27001 and SOC 2; no public trust centre or AI subprocessor list.
• Model disclosure. Providers inside FactSet products are not named.
• Pricing. Not published; AI products are separate SKUs moving toward consumption pricing.
• Expert content. GLG transcripts for mutual clients only.