DAA-C01 Question 15
Select 21.2 Perform data discovery to identify what is needed from the available datasets.A retail analytics team is preparing a new executive dashboard in Snowflake to compare online and in-store sales by region, product category, and promotion. The team has inherited dozens of tables across RAW, CURATED, and FINANCE schemas, but the documentation is incomplete and several columns have unclear names such as SRC_CD, CHNL_FLG, and PROMO_KEY. Before building the semantic layer, the lead data analyst wants to perform data discovery to determine which datasets contain the required business attributes and whether those datasets are suitable for analysis. Which TWO actions should the analyst take first?
- A
Query INFORMATION_SCHEMA and/or ACCOUNT_USAGE views to identify candidate tables, columns, and object metadata, then profile likely columns with targeted SQL summaries such as distinct counts, null rates, and sample values.
- B
Create a materialized view for each table in the RAW schema so that analysts can search the data faster while determining which columns are needed.
- C
Use DESCRIBE, SHOW, and selective exploratory queries on candidate tables to validate grain, key relationships, and whether coded fields map to needed business concepts such as channel and promotion.
- D
Immediately join all sales-related tables from RAW, CURATED, and FINANCE into a single wide table and let dashboard users determine which attributes are useful.
- E
Clone the full database into a new environment and begin transforming all similarly named columns into standardized names before checking whether they are actually needed.
Show answer and explanation
Correct answers: A, C
Explanation
The best initial approach to data discovery in Snowflake combines metadata inspection with targeted profiling of candidate datasets. Analysts commonly start by searching metadata in INFORMATION_SCHEMA or SNOWFLAKE.ACCOUNT_USAGE to find relevant tables and columns, then use SHOW and DESCRIBE commands plus exploratory SQL queries to confirm data grain, business meaning, completeness, uniqueness, and joinability. This aligns with best practices for identifying which available datasets can support required business questions before building semantic models or dashboards. By contrast, prematurely creating materialized views, wide denormalized tables, or broad transformations introduces cost and risk before dataset suitability has been established. Relevant Snowflake references include documentation for INFORMATION_SCHEMA, SNOWFLAKE.ACCOUNT_USAGE, SHOW commands, DESCRIBE commands, and general SQL profiling techniques used to inspect nulls, distinct values, and sample data.
- A. Correct.
Correct. For data discovery, a strong first step is to use Snowflake metadata sources such as INFORMATION_SCHEMA views and ACCOUNT_USAGE views to locate relevant objects and understand available columns. After narrowing candidates, the analyst should profile the data directly with SQL to inspect completeness, cardinality, and representative values. This helps determine whether a dataset actually contains the required business attributes and whether it is fit for analysis.
- B. Incorrect.
Incorrect. Materialized views are designed to improve performance for repeated query patterns, not to support initial discovery of unknown datasets. Creating materialized views across RAW tables would add unnecessary cost and operational complexity before the analyst even knows which datasets are relevant.
- C. Correct.
Correct. DESCRIBE and SHOW commands are useful for inspecting object definitions, while exploratory SQL on a subset of candidate tables helps validate data grain, key structure, and encoded fields. For example, checking whether CHNL_FLG maps cleanly to online versus store channels or whether PROMO_KEY links to a promotion dimension is exactly the type of discovery work needed before semantic modeling.
- D. Incorrect.
Incorrect. Building a wide table before understanding source meaning, grain, and quality is risky and contrary to good analytical design. It can create duplicate facts, incorrect joins, and misleading metrics. Data discovery should come before broad consolidation.
- E. Incorrect.
Incorrect. Zero-copy cloning is useful for safe experimentation, but cloning and mass-standardizing columns is not an efficient first action for identifying what data is needed. The analyst should first discover which datasets and fields are relevant; otherwise, they may spend time transforming unnecessary objects.