COF-C03 Question 133
Single answerCortex AnalystA retail analytics team wants business users to ask questions such as "What were online sales in the Northeast last quarter?" and receive reliable SQL-backed answers through a natural language interface. The source data is spread across several fact and dimension tables with business-specific terminology such as "online sales," "Northeast," and "last quarter." The team wants to improve answer accuracy and reduce ambiguity before exposing the experience to end users. Which action should they take when implementing Snowflake Cortex Analyst?
- A
Create a semantic model that defines the relevant tables, relationships, metrics, dimensions, and business synonyms for the domain
- B
Convert all source tables into a single large denormalized table so Cortex Analyst can only work with one table at a time
- C
Create a Cortex Search Service on the sales tables because Cortex Analyst requires vector search to generate SQL answers
- D
Enable cross-region database replication so Cortex Analyst can interpret time periods and business terms consistently
Show answer and explanation
Correct answer: A
Explanation
The best answer is to create a semantic model for Cortex Analyst. In practice, Cortex Analyst relies on a curated semantic layer to understand how business users refer to data and how that language maps to Snowflake tables, joins, dimensions, and metrics. This is especially important when multiple related tables are involved and when users use business-friendly terms instead of physical column names. A strong semantic model improves the reliability of generated SQL and reduces ambiguity. By contrast, Cortex Search Service serves a different purpose, and infrastructure features such as replication do not solve semantic interpretation problems. This aligns with Snowflake guidance that Cortex Analyst is used for natural language querying of structured data through a semantic model that captures business meaning.
- A. Correct.
Correct. Cortex Analyst is designed to answer natural language questions over structured data by using a semantic model. Defining tables, joins/relationships, measures, dimensions, and business terminology helps the service map user phrasing like "online sales" or "Northeast" to the correct underlying SQL logic. This is the primary way to improve answer quality and reduce ambiguity in real implementations.
- B. Incorrect.
Incorrect. Although denormalization can sometimes simplify BI workloads, Cortex Analyst does not require all data to be flattened into a single table. In many real environments, preserving a logical star or relational model and expressing that structure in the semantic model is the better practice. The misconception is assuming the service cannot reason across multiple related tables.
- C. Incorrect.
Incorrect. Cortex Search Service is a separate capability for low-latency search and retrieval over text or semi-structured content. Cortex Analyst does not require Cortex Search Service in order to interpret questions and generate SQL over structured tables. This option reflects confusion between search/retrieval use cases and semantic SQL generation.
- D. Incorrect.
Incorrect. Cross-region replication is used for disaster recovery, business continuity, and data distribution scenarios. It does not define business semantics such as custom metrics, geographic mappings, or time-intelligence concepts for Cortex Analyst. Replication would not address the core problem of translating business language into accurate SQL.