Use DataAI ETL with Power BI

Execution model: DataAI runs before Power BI in customer Spark/Databricks compute and publishes governed tables or views. Power BI visualizes those outputs through a native customer connector. No DataAI .mez connector is included or required.

1. Produce BI-ready DataAI outputs

Run the installed DataAI quality pipeline or Java/Spark functions.
Persist only approved outputs in a SQL-accessible catalog: run summaries, findings, field profiles, clean/rejected summaries, analytical results, and matrix outputs.
Add run_id and completion time to multi-run result tables.
Expose governed views that omit sensitive source values unless record-level BI access is authorized.
Grant the Power BI identity read-only access to those views.

2. Connect Power BI

In Power BI Desktop, select Get data.
Choose the customer's supported Databricks, Spark, Fabric, SQL, or database connector.
Enter the customer-managed server/warehouse details and approved authentication method.
Select Import or DirectQuery according to the customer's scale, latency, and security requirements.
Load the approved DataAI tables/views.

3. Build the model

Output	Recommended Power BI use
Pipeline runs / dashboard metrics	Quality score, read/accepted/rejected counts, duration, and trend cards
Quality findings	Severity, rule, field, finding-code, and remediation analysis
Field profiles	Completeness, distinct counts, ranges, and null trends
Analytical function results	Ranking, variance, correlations, market, time-series, map, alerts, and narratives
Matrix balance outputs	Original versus balanced values, error, iteration, and convergence visuals

Relate run-level tables to findings and profiles by run_id. Relate findings to source-level rows by record key only when explicitly authorized.

4. Refresh and secure

Publish to an approved Power BI workspace.
Configure the gateway only when the connector and network path require one.
Store credentials in Power BI's credential facility.
Configure refresh after the DataAI Spark job completes.
Apply row-level security, privacy levels, sensitivity labels, and workspace permissions.
Test schema changes, empty runs, failed DataAI jobs, late outputs, and query performance.

Power BI does not execute DataAI functions. Refreshing the semantic model does not rerun Spark unless the customer explicitly orchestrates both activities.

Usage is successful when: Power BI reads only governed outputs, refresh follows a completed DataAI run, relationships and measures reconcile, and security prevents unauthorized record-level access.