Prepare and test your skills

Prepare and test your skills

Establishment of a single source of truth via LookML means everyone in an organization uses the same centrally defined business rules when asking questions of data. Looker (opens in a new tab) uses LookML as a semantic layer between business users and the database, so a metric such as revenue is defined once and reused in every Explore, Look, and dashboard. Because all queries flow through those shared definitions, different teams receive consistent and accurate results instead of conflicting numbers.
A LookML project is the collection of files that defines how Looker connects to data and carries out SQL queries. The LookML project describes the database connection to use, how tables relate to each other, and which options appear in the Looker user interface. Each LookML project lives in its own Git repository for version control, so changes are tracked, reviewed, and deployed in a controlled way. Once the database connection ties the project to one governed data source, an update in one place flows to every query that depends on the project.
A LookML project typically contains four kinds of shared definitions:
Models, views, and Explores in Looker work together to force business users to query through the same rules. A model contains Explore and join definitions, so it controls which tables can be queried together and how the joins behave. A view contains dimensions, measures, and field sets, so it controls how raw columns are named, grouped, and calculated for business use. An Explore combines a model and its views into a query interface, so users select fields from the Explore without writing their own joins or SQL.
System architecture showing a governed database flowing through a LookML project where a model supplies joins and views supply dimensions and measures to an Explore, so business users, Looks and dashboards all get consistent results.
This structure creates a clear dependency chain for accuracy. The Explore depends on the model for join paths, the model depends on the database connection for access, and the Explore depends on views for field definitions. When a user builds a question, Looker combines the model join logic with the view field definitions to generate SQL, then sends that SQL through the database connection to the database. Looks and user-defined dashboards contain no LookML of their own, so their queries also rely on the same underlying LookML elements.
Exam tip: Define a metric once as a measure in a view, not separately in each dashboard, so every Look and dashboard that uses that view returns the same result.
Empowerment through self-service exploration and visualization means non-technical users can independently ask questions of data and build visualizations without writing SQL. Looker (opens in a new tab) places a centralized semantic model written in LookML between users and the database. The semantic model defines business terms, metrics, joins, filters, aggregations, and data permissions, so users work with familiar fields while Looker handles the technical query work.
Looker's semantic model is the centralized LookML layer that lets business users explore data without knowing SQL joins. In an Explore (opens in a new tab), a user selects fields, filters, sorts, and limits, and Looker composes and executes the SQL query using the LookML model. The user never specifies how tables join together, which is one of Looker's most powerful benefits for business users. A Looker query is defined by its model, its Explore, the fields to query and display, the sort field and order, and the filter fields and values.
For example, a user can ask for a count of orders grouped by state and filtered by order creation date. Looker generates SQL that includes the correct join between orders and users, applies the filter, groups the results, sorts them, and limits the rows. Because the model enforces join logic, filtering, aggregation, and data permissions, every self-service query follows the same governed structure.
Conversational Analytics (opens in a new tab) in Looker lets users ask natural-language questions and receive answers based on data. Conversational Analytics uses Gemini for Google Cloud (opens in a new tab) and uses the Looker semantic model as its source of truth, so responses stay accurate and consistent. At the start of a request, Conversational Analytics fetches the schema from the connected Explores and maps words in the question to fields using LookML parameters such as name, label, description, type, and dimension_group. It uses label for user-friendly naming and value_format for formatting, and it ignores fields marked hidden.
Instead of querying the database directly, Conversational Analytics only decides which fields, filters, sorts, and limits should be used. Looker then composes and executes the query using the underlying LookML model, so all queries adhere to the defined join logic, filtering, aggregation, and data permissions. Conversational Analytics can rephrase a question, ask follow-up questions to clarify ambiguity, and remember previous questions and answers within the active conversation. Users can stop a running response, delete the most recent question and response, and view how the query was interpreted and calculated.
Looker offers two types of agents for conversational analytics:
| Agent type | Scope | Entry point and sharing | Limitations |
|---|---|---|---|
| Agents | Specialized standalone agents for business units, cross-domain analysis across up to five connected Explores, authored context with instructions and glossaries, and Python code execution if enabled | Selected from the Agents tab on the Conversational Analytics page, shared with View or Manager access, and publishable to Gemini Enterprise | Requires access_data and other model permissions, sharing does not grant Explore access, and verified queries with pivots or custom fields are not supported |
| Dashboard agents | Natural-language querying limited to a specific dashboard and its query-linked tiles, with custom instructions except on LookML dashboards | Started with Chat with this dashboard, inherits the dashboard sharing model, and conversations appear only from Recent conversations | Cannot be shared independently, do not appear on the Conversational Analytics page, and do not support Advanced Analytics |
Exam tip: Standalone agents can be shared and can use up to five connected Explores, while dashboard agents inherit the dashboard's sharing model and are limited to that dashboard's context.
Visualization tools in Looker turn query results into charts, pivot tables, and interactive dashboards. In SQL Runner (opens in a new tab) visualizations, any numeric field is interpreted as a measure and any non-numeric field as a dimension. Users can override this default from a column's gear menu in the Results area by choosing Convert to Dimension or Convert to Measure. Users pivot results by choosing Pivot Column from the gear menu and unpivot by choosing Unpivot Column on the pivoted column.
SQL Runner visualizations have several limitations:
Totals in Looker follow specific rules about what can be summed. Column totals are available only for measures and table calculations that reference only measures, not for dimensions. Row totals are available only for measures, not for dimensions or table calculations based on dimensions. The Row Totals feature creates an additional query with a limit of 30,000 rows, reduced further by the number of pivot options when data is pivoted. Totals may appear too high when a query is filtered by a measure, uses merged results, or has row or column limits, because totals are calculated before those filters, merges, or limits are applied.
Real-time data access and integrated business actions in Looker means users explore live database data and then use those insights where work happens. Looker (opens in a new tab) uses an in-database architecture, so each question becomes SQL that runs directly against the connected database instead of a stale extract. A centralized semantic model defines business meaning and permissions, so live answers stay consistent for everyone. Specialized agents and dashboard assistants can then be shared or published to external applications, so governed insights trigger follow-up work outside Looker.
Live in-database access in Looker means every Explore, dashboard, or natural-language question is translated into SQL that the connected database executes. Data flows from database tables to the Looker query engine to the results table and visualization, so the answer reflects current data at query time. The user never writes join logic by hand because the model supplies table relationships, filters, and aggregations. SQL Runner (opens in a new tab) provides the same live path for hand-written SQL and for inspecting machine-generated SQL.
A four-step left-to-right flow showing a user request, Looker composing SQL with LookML, live execution in the database, and return of visualizations and dashboards.
The live query flow follows an ordered sequence:
SQL Runner makes this live path visible and testable for beginners. A user can run any valid SQL for the database dialect against a Looker connection, view up to 5,000 rows in the results area, and use error highlighting that shows the location of the first error. Because every Looker query is complete SQL, a user can copy the SQL from the SQL tab, run it in SQL Runner, and confirm identical results.
The central semantic model in Looker is the LookML definitions that serve as the single source of truth for fields, joins, and metrics. The LookML model encodes how the orders and users tables join on orders.user_id = users.id, what terms like revenue or churn mean, and which data permissions apply. Conversational Analytics (opens in a new tab) fetches the LookML schema at the start of a request and maps user words to the correct fields, so non-technical users get governed answers without learning SQL joins. Query generation depends on this model because Conversational Analytics only chooses fields, filters, sorts, and limits, while Looker handles join logic, aggregation, filtering, and permissions.
Conversational Analytics grounds its responses in LookML schema metadata such as:
Integrated business actions in Looker mean governed insights can be used inside external workflows without copying data or logic. Conversational Analytics (opens in a new tab) supports standalone analytics agents for business units and dashboard agents tied to a specific dashboard, and standalone agents can be published to external applications like Gemini Enterprise. The agent carries the same LookML definitions and permissions outward, so teams act on consistent metrics in the tools where decisions are made. Managing LookML files and Git with version control (opens in a new tab) keeps those shared definitions controlled as they are reused externally.
| Agent type | Scope of analysis | How access and sharing work |
|---|---|---|
| Analytics agent | Specialized assistant for a business unit with cross-domain analysis across up to five connected Explores, plus authored context and Python code execution if enabled | Shared by granting content access with View or Manager access, and publishable to Gemini Enterprise, but sharing does not grant data access |
| Dashboard agent | Natural-language assistant limited to one dashboard and its query-linked tiles, with custom instructions except on LookML dashboards | Inherits the dashboard sharing model through Manage access and model permissions, cannot be shared independently, and conversations appear only from Recent conversations |
The sharing boundary protects live data when insights move into workflows. Sharing an analytics agent does not grant access to its underlying Explores, and each user still needs access_data and other model permissions to see results. Dashboard agents enforce the same idea more strictly because access is managed only through dashboard-level and model permissions.
Exam tip: Sharing an agent never grants data access — the recipient still needs permissions on the underlying models and Explores to run live queries.
LookML creates a single source of truth in Looker by defining business rules and metrics such as revenue once in a centralized semantic layer that every Explore, Look, and dashboard reuses. Because all queries flow through those shared definitions, different teams receive consistent and accurate results instead of conflicting numbers.
A model in Looker controls which tables can be queried together and how the joins behave, a view controls how raw columns are named, grouped, and calculated as dimensions and measures, and an Explore combines them into a query interface. Users select fields from the Explore without writing their own joins or SQL, while Looker combines the model join logic with the view field definitions to generate SQL.
Standalone analytics agents in Looker support cross-domain analysis across up to five connected Explores and can be shared with View or Manager access or published to Gemini Enterprise, while dashboard agents are limited to natural-language querying of a specific dashboard and its query-linked tiles and inherit that dashboard sharing model. Sharing an agent never grants data access, so each user still needs access_data and other permissions on the underlying models and Explores to see results.
A retail company wants to modernize its business intelligence and analytics operations. The organization has the following technical and operational requirements:
Which Looker architecture and feature set satisfies these requirements?
Last updated