Move selected Power BI dashboards to Metabase
Assess selected Power BI dashboards for a Metabase rebuild using metric definitions, DAX gaps, permission tests and a synthetic reconciliation dataset.
Last updated
In short
- This is a dashboard rebuild, not an automatic PBIX conversion.
- Validate metric semantics under filters before comparing chart appearance.
- Row-level access and scheduled delivery need separate edition-aware acceptance.
On this page (7)
Migration plans · Part of Metabase
Four fictional orders with known department and total net-sales expectations.
Download the synthetic reconciliation datasetMove suitable analytical workloads deliberately
A Power BI report combines data connections, transformations, model relationships, calculations, visual interactions and access rules. Moving selected workloads to Metabase is generally a redesign and rebuild of those elements, not a direct conversion of a PBIX file.
This is a reference assessment and synthetic worked example. It is not a tested conversion tool or a claim of feature parity. Keep Power BI for reports whose required semantics or workflows have not been reproduced and accepted. A mixed BI estate can be a sensible migration outcome.
Inventory the report’s actual dependencies
| Power BI element | Target treatment | Acceptance question |
|---|---|---|
| Source data connection | Reconnect through a supported Metabase driver | Same records and freshness? |
| Power Query transformations | Rebuild in a reviewed data pipeline or SQL layer | Same types, joins and exclusions? |
| Model relationships | Recreate the intended analytical joins | Same grain and cardinality? |
| DAX measures | Translate business meaning into suitable target logic | Same result under every relevant filter? |
| Visuals and interactions | Rebuild supported charts and navigation | Users can answer the required question? |
| Row-level security | Redesign using supported permission mechanisms | Unauthorized user cannot query or export restricted rows? |
| Subscriptions and embedding | Reconfigure with edition review | Same approved recipients and access boundary? |
Microsoft’s measure documentation explains that measures respond to report interaction and filter context. Translating a formula’s text without reproducing that context can produce a plausible but incorrect metric.
Start with a simple, valuable report
Choose a report with a clear owner, a small number of well-defined measures and a supported data source. Avoid starting with the most complex semantic model, custom visual or embedded application. Record the Power BI model/report revision, Metabase version and edition, database version and source-data snapshot used for comparison.
Define the business metric in ordinary language before writing SQL. For example: net sales means completed order value minus returns, excluding cancelled orders, grouped by the organization’s reporting date. Clarify whether tax, shipping and exchange-rate adjustments belong in that definition.
Use a synthetic reconciliation example
The downloadable fixture contains four fictional orders. It is an author-created test dataset, not customer data or a measured migration result.
| Order | Department | Gross | Returns | Status |
|---|---|---|---|---|
| A100 | North | 100 | 10 | completed |
| A101 | North | 50 | 0 | cancelled |
| A102 | South | 200 | 20 | completed |
| A103 | South | 80 | 0 | completed |
Under the stated definition, North net sales is 90, South is 260 and the total is 350. The cancelled order contributes nothing. These expected values can be checked in both tools before adding production complexity.
An illustrative PostgreSQL query for a pilot table is:
SELECT department, SUM(gross_amount - return_amount) AS net_sales
FROM migration_fixture.orders
WHERE status = 'completed'
GROUP BY department;
The table and column names are a reference schema, not a supplied production migration script. Metabase’s SQL editor documentation describes native queries. Add filter parameters through the supported mechanism rather than concatenating untrusted SQL.
Reconcile semantics, not just totals
Test department filters, date boundaries, empty selections, null values, negative amounts and a period containing no orders. Compare subtotals and grand totals because some measures are not additive. Ratios, distinct counts and time-intelligence calculations can behave differently from a sum.
For every metric, retain its business definition, source formula, target expression, fixture and expected outputs. Have the report owner approve discrepancies or redesign decisions. A matching top-line total can hide incorrect row-level behavior or a join that duplicates some records while excluding others.
Use a representative frozen production sample only after the synthetic cases pass and access is approved. Reconcile record counts, key sets, date ranges and important aggregates. Record timezone and decimal-rounding rules so both tools evaluate the same facts.
Redesign permissions and delivery
Metabase’s data-permission documentation describes available controls and their plan boundaries. Check the selected edition for row/column restrictions, impersonation or other advanced mechanisms. Do not assume Power BI’s security roles transfer through a datasource connection.
Test a North-only user against dashboard filters, direct questions, SQL access, exports, public links and subscriptions. A dashboard filter is a usability feature unless the permission system independently enforces it. The read-only PostgreSQL connection guide covers the database credential boundary, which remains separate from end-user data restrictions.
Recreate scheduled delivery only after recipient lists and data scope are approved. Keep sensitive reports out of unrestricted public sharing during the pilot.
Cut over by report, with a retained baseline
Run both reports against the same refresh window for an agreed observation period. Record discrepancies and resolve them before changing the authoritative dashboard link. Train users on changed interactions and identify any retained Power BI workflow, including exports or embedded experiences.
For rollback, keep the original report, model, data pipeline and necessary subscription active through the acceptance window. Disable duplicate scheduled messages before switching back. Preserve any new target definitions and user feedback rather than deleting the pilot.
Back up Metabase’s application database and the reviewed SQL/model definitions. Retire a source report only after its owner accepts metric fidelity, permissions, delivery and recovery. The right success measure is a reliable analytical workflow, not the number of charts recreated.
Sources and further reading
Use the linked documentation to confirm the version, edition and contract that apply to your project.
