Anthropic recently reported that Claude now handles around 95% of its internal analytics requests, letting employees query business data independently instead of relying on data teams. The company attributes this result less to advances in models and more to data governance, semantic definitions, and operational discipline.
The report argues that AI analytics is only as reliable as the underlying data platform, making data modeling, testing, metadata management, and quality checks critical to accuracy. Chen Chang, Clement Peng, Justin Leder, Johanne Jiao, and Josh Cherry write:
At Anthropic, 95% of business analytics queries are automated via Claude, with ~95% accuracy in aggregate. By giving this often rote, repetitive work to Claude, our data science team can focus on more strategic work like causal modeling, forecasting, and machine learning.
Anthropic highlights how AI analytics depends less on the model itself and more on a small set of governed, canonical datasets, enforced standards, centralized data artifacts, and well-maintained metadata. These reduce ambiguity, prevent metric drift, and help AI systems locate the correct data and definitions.
The authors report that Claude answered only 21% of analytics questions correctly without skills. After encoding analytical workflows and business context as skills, accuracy rose to more than 95% overall and approached 99% in some domains.
Anthropic’s approach tackles a common analytics challenge: self-service access can create overlapping datasets and conflicting metric definitions, while tightly scoped reporting environments often fail to support long-tail business questions and lead to dashboard proliferation. The five members of the data science and data engineering team write:
If data foundations are the data warehouse itself, sources of truth are the reference surfaces the agent consults to navigate it. This layer reduces concept <> entity ambiguity and turns "weekly active users" in a stakeholder’s question into a specific, governed entity in your data model. The analytics setup is built on four layers: data foundations (governed models, metrics, and metadata), the knowledge layer (semantic definitions, lineage, and business context), the skills that encode repeatable analytical workflows, and validation systems that verify outputs for correctness and consistency.
Source: Anthropic blog
The company concludes that successful AI analytics depends on three principles: maintaining a single source of truth for metrics, making the right data easy to find, and continuously detecting stale definitions. The reaction from the data community has been mixed, with some highlighting openness and others arguing that analytics should produce deterministic, idempotent results. Francesco Mucio, owner and BI/data architect at Untitled Data Company, writes:
I saw a lot of not-so-good takes on the Anthropic article about how they do self-service analytics. I am going to make it very clear. How do they do it? With a Semantic Layer. The data foundation is a dimensional model, but they don't query the tables directly, they use the analytics skills to first use the Semantic Layer to figure out dimensions, metric definitions, and joins.
Anthropic considers semantic metrics, lineage, query patterns, and business context to be the key sources of truth for analytics agents. Structured definitions matter more than raw query history, while human-owned documentation remains essential. Arsenii Antonenko, automation QA engineer, comments:
It's interesting that more and more real-world deployments point to the same conclusion: AI performance is often constrained less by model capability and more by context definition.
An appendix to the article provides a redacted template of the skill file used to guide analytics agents.