DBX-DAA Study Guide: One Business Question, Five Places to Answer It, and Whose Permissions Apply at Each
A sales manager asks: "What was revenue by region last week?" On Databricks an analyst can answer that in at least five places: a query in the SQL editor, a view, a materialized view, an AI/BI dashboard, or a Genie Agent the manager asks directly. The Databricks Certified Data Analyst Associate exam is largely about choosing between them, and about two details that change at each step: which compute runs the query, and whose data permissions apply. This guide follows one question through all five.
Study guide · Databricks Certified Data Analyst Associate (DBX-DAA)

The exam
45 scored multiple-choice questions (plus up to 10 unscored) in 90 minutes. The pass mark is not published; check the official exam page. Nine sections, and four of them carry 63% of the weight:
- Executing Queries using Databricks SQL and SQL Warehouses: 20%
- Dashboards and Visualizations: 16%
- Analyzing Queries: 15%
- AI/BI Genie Agents (Genie Spaces): 12%
- Understanding of the Databricks Data Intelligence Platform: 11%; Managing Data and Securing Data: 8% each; Importing Data and Data Modeling with Databricks SQL: 5% each.
This is a scenario exam. Of the 225 DBX-DAA questions we wrote for this question bank, 52% are use-case questions ("an analyst needs to ... which approach fits?"), and only about a third quote any SQL or code. You need to read SQL fluently, but most points come from picking the right object, setting or feature.
Place 1: a query in the SQL editor
The query runs on a SQL warehouse, the compute for Databricks SQL. Data lives in cloud storage, Unity Catalog governs it, and the warehouse does the work. To read sales.emea.orders the analyst needs USE CATALOG, USE SCHEMA and SELECT; BROWSE only lets them see the table's name and comments in Catalog Explorer, which explains the stem where someone finds a table in search and still gets a permission error.
Questions in this place test SQL that gives a wrong answer without an error. Learn these by heart:
WHERE region = NULLreturns nothing; useIS NULL.- An aggregate in
WHEREfails; the condition belongs inHAVING. - A
LEFT JOINfollowed by aWHEREcondition on a right-side column quietly becomes an inner join; move the condition intoON. - A
SUMthat is too high after a join is fan-out: aggregate the many side to one row per key first. count(col)skips NULLs,count(*)does not.
Genie Code (formerly Databricks Assistant) can write and fix SQL in the editor, but the exam still expects you to know why the fix works.
Beginner tip: make two tiny tables in the SQL editor, customers with three rows and orders where one customer has two orders and one has none. Write the LEFT JOIN, then add WHERE o.amount > 0 and watch the customer without orders vanish. Move the condition into ON and watch it come back. Then sum a customer column after the join and see it double. Those two experiments cover a large share of the "why is this result wrong?" questions.
Place 2 and 3: a view or a materialized view
If the same query will be reused, save it. The choice is about when the cost is paid:
| Object | What it stores | Choose it when |
|---|---|---|
| Standard view | Only the query; every read runs it in full. | Cheap logic that must always be current. |
| Dynamic view | Only the query, using is_account_group_member() or session_user() to hide rows or columns per reader. | Different groups must see different data. Remember: anyone with SELECT on the base table bypasses it. |
| Materialized view | The precomputed result, refreshed manually, on a schedule, by a job, or automatically when source tables change (TRIGGER ON UPDATE). | An expensive aggregation feeding dashboards; reads are fast, cost moves to refresh time. |
A materialized view shows data as of its last refresh. When a stem says the numbers on the dashboard are "a day behind", look at the refresh schedule before anything else. For audits, Delta history helps: DESCRIBE HISTORY tells you who changed a table and when, and VERSION AS OF lets you query the state behind last month's report.
Place 4: an AI/BI dashboard
A dashboard has a Data tab with datasets and a Canvas with widgets. Shape the dataset once (needed columns, needed rows), then let widgets slice it. Two settings decide most dashboard questions.
Filters versus parameters. A field filter narrows the dataset result by a column, needs no change to the SQL, and on small datasets runs in the browser. A parameter such as :start_date changes the SQL text itself, so it always runs on the warehouse. Use a parameter when the value must reach inside the query, for example a date in a subquery.
Publish settings. With shared data permissions (the default) every viewer sees what the publisher can see, so viewers need no access to the tables. With individual data permissions queries run as each viewer, and their own row filters and column masks apply. Compute comes from the publisher either way. A stem that says "regional managers must only see their own region" wants individual permissions or a row filter, not separate dashboards.
Going deeper: why the second run of a query is fast, and when it is not
Databricks SQL has several caches. The result cache returns the stored result of an identical query on unchanged tables almost at once. Its local part lives in the warehouse's memory and is lost when the warehouse stops; on serverless warehouses a remote result cache (for ODBC/JDBC clients and the SQL Statement API) also survives a stop. The disk cache keeps data files read from storage on local SSDs and is cleared on stop as well. That is why the first heavy query after an overnight auto-stop is slow and the next one is fast, and why timing a rewrite fairly needs SET use_cached_result = false. Dashboards add their own behaviour: small datasets are filtered in the browser after the first load, so those interactions do not appear in query history at all.
Place 5: a Genie Agent
Genie Agents (formerly Genie spaces) let the manager ask the question in plain language. The exam tests the setup and, again, the permissions:
- Compute and data use different identities. Saving a default warehouse embeds the author's compute credentials, so users need no warehouse permission. Data access is always checked with the asking user's own Unity Catalog permissions. Sharing an agent does not grant data.
- Context, in order of trust. Attached tables and metric views with good descriptions first, then example SQL queries and SQL expressions for measures and filters, and plain-text instructions last. Start small: a narrow purpose and few tables.
- Measuring quality. Benchmarks with a SQL Answer are scored automatically; user feedback ("Is this correct?") shows editors where to add examples but never changes the agent by itself.
Going deeper: why "Florida" can stop matching
Genie writes SQL against stored values. When a user types "Florida" and the column stores FL, prompt matching (format assistance and entity matching) bridges the gap using a stored list of the column's distinct values. That list is a snapshot. When new codes or formats arrive in the data, the snapshot is stale and Genie starts producing filters that match nothing. The fix questions point to is Refresh prompt matching on that column, not a new instruction.
Where the remaining points sit
Importing Data: read_files to explore files without creating anything, the upload UI for a one-off file into a volume, OpenSharing (formerly Delta Sharing) and Marketplace for shared data that arrives as a read-only catalog, and Lakehouse Federation to query an external database in place. Data Modeling: star schema facts and dimensions, and the medallion layers. Analyzing Queries: Query History and the query profile, Photon, Liquid Clustering and predictive optimization on managed tables.
A study order
- SQL correctness: joins, aggregates, NULLs, window functions, set operators.
- SQL warehouses, views versus materialized views, managed versus external tables.
- Dashboards: datasets, filters, parameters, publishing, schedules, alerts.
- Genie Agents, then query analysis and caching.
- Platform, importing, modeling and securing data, then mixed sets.
Practise
Practise in the question bank: every question explains each option, right and wrong. Start with the domain you just studied, then take a timed mock exam.
- DBX-DAA exam page · All 285 DBX-DAA questions · Timed mock exam · 17-day study plan
- By domain: Understanding of the Databricks Data Intelligence Platform · Managing Data · Importing Data · Executing Queries using Databricks SQL and SQL Warehouses · Analyzing Queries · Dashboards and Visualizations · AI/BI Genie Agents (Genie Spaces) · Data Modeling with Databricks SQL · Securing Data
Section weights as listed in the Databricks exam guide for the Data Analyst Associate. Product names in this area change often (Genie spaces are now Genie Agents, Databricks Assistant is now Genie Code); check the current exam guide and documentation.