ARA-C01 Question 195
Single answerSnowflake ML functionsA retail company stores customer support tickets in Snowflake and wants to classify each ticket into one of several issue categories directly in SQL, without moving data to another platform. The architecture team wants the fastest path to production with minimal custom model training or infrastructure management. They also need the solution to work inside Snowflake security boundaries and be callable from SQL pipelines. Which approach best meets these requirements?
- A
Use Snowflake Cortex classification capabilities from SQL to perform managed text classification on the ticket text inside Snowflake.
- B
Export the ticket data to an external ML platform, train a custom classifier there, and write predictions back to Snowflake through batch loads.
- C
Create a materialized view on the ticket text and rely on automatic query optimization to infer issue categories during query execution.
- D
Use a JavaScript stored procedure to manually implement a Naive Bayes classifier from scratch so the model remains inside Snowflake.
Show answer and explanation
Correct answer: A
Explanation
The best answer is to use Snowflake's managed AI/ML SQL functionality for text classification inside Snowflake. This approach best satisfies the scenario's core architectural constraints: keep data within Snowflake, invoke capabilities directly from SQL-based pipelines, and avoid the overhead of building and operating a custom ML platform. In Snowflake best practices, managed ML/AI capabilities are preferred when the use case matches available built-in functions because they reduce operational burden and simplify governance. By contrast, exporting data to external platforms or hand-coding a classifier increases complexity and weakens the stated goal of rapid delivery. Materialized views and query optimization features do not provide predictive or generative classification behavior. Candidates should recognize when Snowflake ML/Cortex functions are the most appropriate architectural choice for in-platform inference use cases.
- A. Correct.
Correct. For a requirement centered on classifying text directly in SQL with minimal operational overhead and no separate training infrastructure, Snowflake's managed ML/AI SQL capabilities are the best fit. Cortex functions are designed to let teams invoke AI functionality from SQL while keeping data in Snowflake, which aligns with the security-boundary and low-management requirements in the scenario.
- B. Incorrect.
Incorrect. This approach can work technically, but it contradicts the stated goals of minimizing infrastructure management and avoiding data movement. Exporting data to an external ML platform introduces operational complexity, governance concerns, and latency in prediction workflows. A candidate might choose this option because custom external ML is common in many architectures, but it is not the fastest or simplest path here.
- C. Incorrect.
Incorrect. Materialized views improve query performance for precomputed results; they do not provide ML classification capabilities or infer semantic categories automatically. This distractor targets the misconception that query optimization features can replace AI or ML functions.
- D. Incorrect.
Incorrect. Although stored procedures can orchestrate logic in Snowflake, implementing and maintaining a classifier from scratch in JavaScript is not the minimal-effort or best-practice approach for managed text classification. It also shifts responsibility for model quality, training logic, and maintenance onto the engineering team. Someone might pick this if they assume keeping code in Snowflake automatically makes it the most suitable architecture, but managed ML functions are more appropriate.