BI Platforms · 4 min read
Databricks opens bulk /importBI migration for Tableau and Power BI
Databricks' /importBI skill in Genie Code now migrates a folder of Tableau and Power BI reports in one run, covering .twb, .twbx, .tds, .tdsx, .pbit and .tmdl files staged in a Unity Catalog volume.
Key takeaways
- Bulk /importBI migration is generally available and processes a folder of reports in a single run.
- Six source file formats supported: .twb, .twbx, .tds, .tdsx, .pbit, .tmdl.
- Playbook runs in three phases and eight steps, from export and staging through review and refinement.
- Each report carries one of three status states: Queued, Running, or Completed.
- Tableau imports create local metric views by default; Power BI imports create Unity Catalog metric views.

Six file formats. One command. Hundreds of reports.
Databricks' /importBI skill inside Genie Code now converts a whole directory of Tableau and Power BI assets into AI/BI Dashboards (the vendor's hosted BI interface) in a single run, and the company describes the feature as generally available in customer workspaces today.
The earlier flow, set out in a previous Databricks post, moved one Tableau or Power BI report at a time. The vendor calls that workable for a pilot and corrosive at scale, since most organizations accumulate hundreds of reports over years, and one-at-a-time processing turns a modernization program into a months-long manual grind. Bulk migration exists to remove that repetition.
What does the bulk migration flow actually do?
The migration agent reads source BI files, generates the corresponding AI/BI dashboards, and maps them to a target catalog and schema the operator names. Files must sit in a directory inside a Unity Catalog volume (governed storage and metadata layer); dragging files into the chat still works only for single-report imports.
The playbook runs in three phases and eight steps: prepare (export reports, stage them in a volume), run (start the bulk import, choose the target catalog and schema, track the batch), and review and refine (review each report, follow the migration plan, polish).
Supported inputs and invocation forms:
- Six file formats: .twb, .twbx, .tds, .tdsx, .pbit, .tmdl
- Tableau workbooks attached to a published data source need the companion .tds or .tdsx in the same folder
- Command form: /importBI <path/to/volume/directory>, issued from a draft dashboard
- Natural-language form: "Migrate every Power BI report in /Volumes/finance/bi/reports to AI/BI Dashboards"
- Batch-wide instructions can be appended, for example "build only the metric views"
What happens once the run starts?
The agent inventories the volume, pairs each Tableau workbook with its companion data source file on its own, asks for confirmation when a pairing is unclear, and prompts for the destination catalog and schema where metric views (measure definitions in the semantic layer) and SQL views will land.
After approval, a status card tracks every report through three states: Queued, Running, and Completed. Operators can stop one report or halt everything still in flight from the same card.
Each report runs as its own background job, so the tab does not need to stay open. Progress remains visible in Genie Code history, and each file name on the status card opens that report's own conversation in a separate tab, where a paused run waits for an answer before resuming. Per report, the agent parses and assesses the workbook, enumerates the semantic layer, creates metric views, builds dashboard pages, and validates the migration.
How much can go into one run?
The post states no ceiling. Databricks directs readers to the public /importBI documentation for current file-count and size limits, and warns those limits can change. No accuracy rates, migration times, or batch maxima appear anywhere in the guidance, and partner-powered AI features must be enabled at both account and workspace level before a run begins.
For a data team, that means batch size has to be confirmed against the documentation before a migration window is booked.
What should a data team fix before the run?
- Read-only MCP connection (Model Context Protocol, a link between an agent and an external tool) to Tableau or Power BI, so the agent checks the live report instead of reading values from an exported file or screenshot
- Upload shared or published data-source definitions: a .tds file for a Tableau published data source, a .pbit for a Power BI shared data model
- Make the underlying data reachable through Unity Catalog tables via Lakeflow Connect, Lakehouse Federation, or existing tables; the sample prompt reads "Use prod.finance.table1, prod.finance.table2 and prod.finance.mv_revenue as the source for this report"
- Add before-screenshots to the chat, saved to a Unity Catalog volume and referenced by path, which Databricks says significantly improves visual and layout fidelity
- Budget for refinement: a second prompt to adjust a chart, filter, or calculation after the first pass is normal
What does "Completed" certify?
Very little. Databricks' own guidance states that "Completed" means the run finished, not that every visual is final, and each report should be validated before its session closes. The safety net is bounded: every Genie Code response can be undone as a single step, and a follow-up prompt can even beautify a dashboard's look and feel.
One asymmetry deserves attention. Tableau imports produce dashboard-scoped (local) metric views by default, while Power BI imports produce Unity Catalog metric views. Teams expecting reuse across dashboards and applications must promote the local views themselves.
Is the throughput gain worth the trade-off?
The arithmetic favors bulk migration. The single-file flow required one command per report; bulk migration requires one per folder, and the batch runs in the background while reviewers work through results.
That gain sits against three unresolved items: undocumented file-count and size ceilings, a manual promotion step for Tableau metric views, and correctness that still depends on a human comparing migrated dashboards against live source data before retiring anything.
community.databricks.com (Original)
More from Marcus Bennett
Show full bio
News editor covering people, launches and day-to-day practice in Analytics Digest is a professional daily at Analytics Digest.
10 articles