AI Feature Scorecard for Analytics and BI Tools

Three AI models (Claude Opus 5.5, GPT-6.1 Sol and Gemini 3.8 Flash, all at high reasoning effort) read the official public documentation of all 15 tools, 28,214 pages, and answered the same 82 questions, quoting the page behind every answer that finds something. Each worked on its own. Each was then shown what the others found across all the tools and asked to re-check any answer it had marked nothing found. Only those answers could change. A tool’s score is the median of the three models’ scores.

Subscores

• Built-in AI (42 questions): what the tool’s own assistant and agents can do.
• External AI (23): what your own agent, such as Claude or ChatGPT, can do through the tool’s MCP server, CLI, API or agent skills.
• Other (17): business context, data handling and residency, and shared-definition and sharing questions.

Scoring

• Yes earns full credit, partly half, no or nothing found none.
• The score averages four equally weighted areas (analysis, building and maintaining assets, context and connections, reliability and controls), so the three subscores do not add up to it.
• A feature asked about both ways earns 85% of its points from the better route and 15% from the other.
• Only what a tool’s documentation says counts, including pages it links to. Code a vendor publishes elsewhere, such as on GitHub, does not count unless the docs point to it.
• Public preview features count the same as generally available ones, except for the MCP server question, where a preview earns partly. Private previews and roadmap items do not count.