SnowPro Specialty: Gen AI Question 64
Single answerSemantic model generationA retail analytics team is using Snowflake Cortex Analyst to let business users ask natural-language questions about sales performance. They generated an initial semantic model from their SALES database, but users report that questions like "What was net revenue by region last quarter?" often return inconsistent SQL because the model confuses gross sales, discounts, and returns across several similarly named columns. The team wants to improve reliability before publishing the model broadly. Which action is the BEST next step?
- A
Regenerate the semantic model after curating the source schema so only relevant tables and columns are included, and add clear business-friendly names and descriptions for measures and dimensions
- B
Increase the warehouse size used by Cortex Analyst so the generated SQL has more compute available to resolve ambiguous business terms
- C
Convert all numeric columns in the source tables to VARCHAR so the model treats them consistently during semantic model generation
- D
Create additional copies of the SALES tables in a new schema so the semantic model generator has more examples of retail data patterns
Show answer and explanation
Correct answer: A
Explanation
For semantic model generation, the most effective way to improve downstream natural-language query accuracy is to curate the input schema and provide clear semantic metadata. In practice, this means exposing only the tables and columns needed for the use case, organizing metrics and dimensions logically, and supplying business-friendly names and descriptions that reduce ambiguity. This aligns with Snowflake guidance for semantic modeling and Cortex Analyst workflows: the semantic layer should reflect business meaning clearly so natural-language questions can be translated into accurate SQL. Compute scaling addresses performance, not semantic ambiguity, and duplicating or poorly typing source data makes the model less reliable rather than more reliable.
- A. Correct.
Correct. Semantic model quality depends heavily on the quality and clarity of the underlying schema metadata and the scope of objects included. If the model includes multiple similar columns for gross sales, discounts, returns, and net revenue without clear semantic definitions, natural-language interpretation can become ambiguous. A best practice is to limit the semantic model to the relevant tables and columns and enrich measures and dimensions with clear, business-friendly names and descriptions so the model can map user intent more accurately.
- B. Incorrect.
Incorrect. Warehouse size affects compute resources for query execution, but it does not fix ambiguity in semantic meaning. If the semantic model cannot clearly distinguish business concepts like net revenue versus gross sales, adding compute will not improve the correctness of generated SQL in a meaningful way.
- C. Incorrect.
Incorrect. Changing numeric business metrics to VARCHAR would make the source data less suitable for analytics and likely break or degrade aggregations, filtering, and metric interpretation. Semantic model generation benefits from well-typed analytical schemas, not from converting measures into text.
- D. Incorrect.
Incorrect. Duplicating tables does not improve semantic understanding and may actually worsen the problem by introducing even more redundant or overlapping objects. Semantic model generation works best when the model is scoped to a curated, high-signal set of data objects with clear metadata.