SnowPro Associate: Platform Question 326
Single answer● CLASSIFY_TEXT functionA customer support team stores incoming email bodies in a Snowflake table named SUPPORT_EMAILS(email_id NUMBER, message_text STRING). They want to use Snowflake Cortex to automatically assign each email to one of these labels: "Billing", "Technical Support", or "Account Management". The team also wants a confidence score returned with the result so they can route low-confidence emails for manual review. Which approach best meets this requirement?
- A
Use CLASSIFY_TEXT(message_text, ['Billing','Technical Support','Account Management']) and review the returned label and score fields.
- B
Use COMPLETE with a prompt that asks for one of the three labels, because CLASSIFY_TEXT cannot return confidence information.
- C
Use SENTIMENT(message_text) and map positive, neutral, and negative results to the three support categories.
- D
Use EMBED_TEXT_768(message_text) and expect Snowflake to automatically convert the vector into one of the three labels with a confidence score.
Show answer and explanation
Correct answer: A
Explanation
The best answer is to use CLASSIFY_TEXT because it is the Snowflake Cortex function intended for assigning text to one of several predefined labels. This is a practical fit for ticket routing, moderation, and categorization workflows. In contrast, COMPLETE is a general text generation function and is less appropriate for controlled classification when Snowflake already provides a dedicated classifier. SENTIMENT measures tone rather than topic, and EMBED_TEXT_768 produces embeddings rather than category predictions. Snowflake documentation for Cortex AI functions describes CLASSIFY_TEXT as the purpose-built function for label-based text classification and supports using the returned output, including score/confidence-style information, in SQL-driven workflows.
- A. Correct.
Correct. CLASSIFY_TEXT is designed for text classification scenarios where you provide candidate labels and the function returns the predicted class. In Snowflake Cortex, the result includes classification output and a confidence-style score that can be used for downstream logic such as manual review thresholds. This directly fits the requirement to classify support emails into a fixed set of categories.
- B. Incorrect.
Incorrect. COMPLETE can sometimes be used to generate labels with prompt engineering, but it is not the best fit when the goal is a structured classification task against a known set of labels. The statement that CLASSIFY_TEXT cannot return confidence information is the key error. CLASSIFY_TEXT is specifically intended for this use case and is preferable to building a free-form prompt for deterministic category routing.
- C. Incorrect.
Incorrect. SENTIMENT analyzes emotional tone, not business-topic category. A message about billing can be positive or negative, and a technical support request can also vary in tone. Mapping sentiment directly to support queues is a category error and would not meet the requirement.
- D. Incorrect.
Incorrect. EMBED_TEXT_768 generates vector embeddings for text similarity and semantic search use cases. It does not by itself produce business labels or confidence scores. Additional modeling or similarity logic would be required, so this option does not satisfy the stated requirement directly.