SnowPro Specialty: Gen AI Question 114
Single answerCortex AnalystA retail analytics team is building a natural-language dashboard that lets regional managers ask questions such as "What were online sales in the Northeast last quarter by product category?" The team wants to use Cortex Analyst so business users receive SQL-generated answers grounded in curated warehouse data, while minimizing incorrect joins and ambiguous metric definitions. Which action is the BEST way to improve answer accuracy before exposing the solution to end users?
- A
Create a well-defined semantic model that specifies the relevant tables, relationships, dimensions, and metrics used by Cortex Analyst.
- B
Increase the warehouse size so Cortex Analyst has more compute available to infer joins and business definitions at runtime.
- C
Store unstructured product manuals in a Cortex Search service and let Cortex Analyst use that index as the primary source for metric generation.
- D
Grant Cortex Analyst broad access to all schemas in the database so it can choose from the maximum number of tables when generating SQL.
Show answer and explanation
Correct answer: A
Explanation
The best answer is to create a strong semantic model for Cortex Analyst. Cortex Analyst is meant for natural-language analytics over structured Snowflake data, and its accuracy depends heavily on how clearly the business data is modeled. In practice, teams should define a governed semantic layer with approved tables, relationships, dimensions, and measures/metrics so the service can generate SQL that reflects business intent rather than guessing across raw schemas. This aligns with Snowflake guidance to curate the structured data context presented to analytical AI experiences. Increasing compute may help latency but not correctness; using Cortex Search addresses a different problem space focused on retrieval from indexed content; and granting overly broad schema access typically harms precision by introducing unnecessary ambiguity.
- A. Correct.
Correct. Cortex Analyst is designed to answer natural-language questions over structured data by generating SQL grounded in a semantic model. Defining tables, joins/relationships, dimensions, and metrics in that model is the primary way to reduce ambiguity and improve the accuracy and consistency of generated queries. In this scenario, curated business definitions such as 'online sales,' 'Northeast,' and time-period logic belong in the semantic layer.
- B. Incorrect.
Incorrect. Warehouse size affects query execution performance, not the business correctness of SQL generation. A larger warehouse may run generated SQL faster, but it does not resolve ambiguous schema design, unclear metric definitions, or incorrect join paths. This option reflects the misconception that compute can compensate for poor semantic modeling.
- C. Incorrect.
Incorrect. Cortex Search is intended for retrieval over indexed content and is not the primary mechanism Cortex Analyst uses to generate accurate SQL over structured warehouse data. Product manuals are unstructured documents and do not solve the problem of defining governed metrics and table relationships for analytical questions. This distractor targets confusion between retrieval-based document search and semantic modeling for structured analytics.
- D. Incorrect.
Incorrect. Broadening access to all schemas usually increases ambiguity and the chance of irrelevant or incorrect table selection. For Cortex Analyst, exposing only the curated data needed for the use case is a better practice than expanding scope. This option reflects a common but risky assumption that more source data automatically leads to better answers.