Your path to agentic analytics
Follow the road from first principles to deploying a reliable agent. Twelve short lessons, in order.
0 / 12 lessons
That is the whole path. Twelve lessons, first principles to production. Come back to any of them any time.
ReferenceGlossaryEvery term in agentic analytics, defined in one line. 14 terms.
- Agent evaluationAgent evaluation is testing an analytics agent against questions where you already know the right answer, so you can measure its accuracy before you trust it.
- Agentic loopThe agentic loop is the cycle an analytics agent runs to answer a question, planning a query, running it, checking the result, and correcting itself until the answer holds up.
- Business intelligenceBusiness intelligence, or BI, is the practice of turning company data into dashboards and reports. Agentic analytics is a newer approach that answers ad hoc questions instead of pre-building views.
- Context engineeringContext engineering is the practice of building and governing the structured context, the schemas, metrics, rules, docs, and tools, that makes an analytics agent reliable.
- Conversational analyticsConversational analytics lets people ask data questions in a chat interface and get answers back, usually powered by an analytics agent behind the chat.
- Data governanceData governance is the set of rules and controls over who can see and use which data. For an analytics agent it decides what tables and rows the agent is allowed to touch.
- Data warehouseA data warehouse is the central database where a company stores its analytics data. It is the source an analytics agent queries to answer questions.
- HallucinationA hallucination is when an AI model produces a confident answer that is wrong or made up. In analytics it shows up as an invented metric definition or a query that quietly returns the wrong number.
- Metrics layerA metrics layer is where a company defines its key metrics once, so revenue or active users mean the same thing in every report, dashboard, and agent answer.
- Natural language queryA natural language query is a data question asked in plain words instead of SQL, which an analytics agent translates, runs, and answers.
- Retrieval augmented generationRetrieval augmented generation, or RAG, gives an AI model relevant information at query time instead of relying on what it memorized. In analytics it pulls in the right schema, metrics, and docs for each question.
- Self-serve analyticsSelf-serve analytics means business users get their own answers from data without waiting on the data team. Agentic analytics is the latest attempt to make it real.
- Semantic layerA semantic layer is a shared definition of your metrics and dimensions that sits between raw tables and the people or agents asking questions, so revenue means one thing everywhere.
- Text to SQLText to SQL turns a plain-language question into a SQL query in one step. It is a building block of agentic analytics, not the whole thing.