SnowPro Specialty: Gen AI Question 158
Single answerSuggested QuestionsA retail company built a Cortex Analyst-powered chat experience on top of a semantic model for sales data. After a user asks an initial question, the product team wants the UI to present follow-up prompts that help users continue the analysis, such as drilling into regions or comparing time periods. They also want to minimize application-side logic and keep the suggestions grounded in the current analytical context. Which approach best meets these requirements?
- A
Call the Cortex Analyst Suggested Questions capability after the current interaction so it can generate context-aware follow-up questions based on the active conversation and semantic model.
- B
Precompute a fixed list of generic follow-up questions in a lookup table and display the same set after every user query, because Suggested Questions are intended only for model training workflows.
- C
Use AI_COMPLETE directly against the warehouse tables to generate arbitrary prompts, because Cortex Analyst does not support contextual follow-up question generation.
- D
Add hard-coded dashboard filter recommendations in the front end, because Suggested Questions require custom vector embeddings and cannot use the semantic model context.
Show answer and explanation
Correct answer: A
Explanation
The best answer is to use Cortex Analyst's Suggested Questions capability because it is intended to improve the end-user analytics experience by proposing relevant follow-up questions in context. In a real implementation, this helps teams reduce custom front-end logic and maintain alignment with the semantic model rather than generating uncontrolled prompts from raw tables. Static lists and hard-coded recommendations are easier to implement, but they do not adapt to the user's analytical path. Using a general-purpose LLM function such as AI_COMPLETE can work for custom experiences, but it is not the best fit when Cortex Analyst already provides built-in, context-aware follow-up suggestions tied to the analytics workflow. This follows Snowflake best practices of using higher-level Cortex capabilities when they match the use case, especially for governed, semantic-model-based analytics experiences.
- A. Correct.
Correct. The Suggested Questions capability in Cortex Analyst is designed to produce relevant next-step questions for the user based on the current analytical context. This aligns with the requirement to keep suggestions grounded in the conversation and semantic model while reducing custom application logic.
- B. Incorrect.
Incorrect. A static lookup table does not adapt to the user's current question or results, so it fails the requirement for context-aware follow-up suggestions. It also reflects a misconception that Suggested Questions are only related to offline training or curation rather than runtime user assistance.
- C. Incorrect.
Incorrect. AI_COMPLETE is a general LLM function, but using it directly against warehouse tables would require significantly more prompt engineering, guardrails, and application logic to maintain context and semantic consistency. The statement that Cortex Analyst does not support contextual follow-up question generation is false.
- D. Incorrect.
Incorrect. Hard-coded UI recommendations may be useful in some applications, but they do not satisfy the need for dynamic, context-aware suggestions. This option also incorrectly assumes Suggested Questions depend on custom vector embeddings instead of leveraging the existing Cortex Analyst semantic context.