Databricks Power BI integration puts shared metrics ahead of local definitions
Databricks announced a private preview on November 10, 2026, bringing centrally defined business metrics into Power BI while keeping definition changes upstream. The Databricks Power BI integration connects reporting to Unity Catalog, its data governance catalog, through published metric definitions. The promise is shared business meaning across applications; the constraint is that publishing remains directional.
That distinction matters for organizations that want familiar reports without maintaining separate interpretations of revenue, margin, or customer activity. A semantic layer, which defines business metrics and relationships above raw data, determines what those numbers actually mean. Databricks wants that authority upstream of individual reporting tools.
Microsoft also positions Power BI semantic models, which package business calculations and relationships for analysis, as authoritative business resources. The immediate choice for customers concerns ownership: which system defines a metric, and which systems consume it? This preview offers a concrete answer for Databricks-centered teams, while leaving synchronization, security, and compatibility work on their side.
What the Databricks Power BI integration changes
The preview changes where reporting definitions originate, while preserving Power BI as a destination for business users.
According to the metric publishing announcement, organizations can publish Unity Catalog metric views to Power BI. Metric views are reusable definitions of business measurements and analytical dimensions. Each published view becomes a model table, with dimensions represented as columns and eligible measures translated into DAX, Power BI’s calculation language.
The announcement describes support for standard aggregations, ratios, distinct counts, and selected window measures, which calculate across defined time or ordering ranges. It also describes publication through Catalog Explorer, the catalog interface, or automated jobs. Access remains in private preview through a Databricks account contact, rather than general availability.
The practical change is a publication path between environments that otherwise invite teams to maintain parallel definitions. A reporting specialist can work with recognizable fields and measures without independently recreating every approved calculation. That reduces a particular source of disagreement: independently maintained implementations of the same business rule.
Consider an illustrative revenue review. Finance excludes canceled transactions, while a sales dashboard counts orders when they enter the system. Both reports can accurately query their underlying data and still disagree. The disagreement comes from definitions, rather than a broken chart or an arithmetic error.
Central authorship gives those teams somewhere specific to resolve the disagreement. They must decide whether the measurements describe the same business concept. If they do, they can share an approved definition. If they do not, the organization should name them differently and explain their separate purposes.
Publication does not make that business decision. It carries the decision into another application after someone has made it. That boundary keeps the announcement grounded: the integration addresses duplicated implementation, while organizational agreement remains a prerequisite.
For readers evaluating the Databricks Power BI integration, this is the meaningful event. Databricks is offering a route for existing reporting workflows to consume definitions maintained elsewhere. Customers can evaluate that route without treating a reporting interface migration as the central project.
The preview should therefore be judged by whether it reduces reconciliation work for actual reports. A successful demonstration includes agreement about the meaning of a number, alongside agreement about its displayed value. Matching screenshots alone provide weaker evidence.
Shared definitions move accountability upstream
Centralizing metric authorship makes data teams responsible for the downstream consequences of definition changes.
The pressure falls most directly on teams that maintain Databricks data products while supporting established Power BI reporting. Shared definitions offer them fewer parallel implementations to reconcile. In exchange, their changes become dependencies for colleagues whose reporting schedules and priorities differ.
A department-owned calculation can change within a relatively narrow working group. A shared revenue definition requires broader coordination because several consumers depend on its meaning. The technical ability to publish a change does not establish who should approve it.
That creates an ownership question before it creates a tooling question. Finance might own the accounting interpretation, data engineering might implement it, and reporting teams might manage presentation. A central definition works best when those responsibilities are explicit enough to resolve disagreements.
For an illustrative campaign report, analysts might compare spending with attributed revenue. A change in attribution policy would alter the interpretation of campaign performance even if the underlying transactions stayed unchanged. Business users need to understand that change before comparing results across reporting periods.
The central model owner would therefore need to distinguish a correction from a policy revision. A correction fixes an implementation that failed to express the existing rule. A policy revision changes the approved rule itself. Those events deserve different explanations, even when both arrive as updated calculations.
Power BI semantic models remain important in this arrangement because they sit close to the reporting experience. Their owners understand report layouts, user expectations, and dependencies that upstream teams might overlook. Central authorship should give those owners a clearer dependency to manage.
An upstream team that treats publication as the end of its responsibility would miss this operational shift. A useful definition includes an accountable owner, an explanation, and a process for handling questions. Without those elements, a shared calculation can become a centralized source of confusion.
The same issue applies to AI-assisted analysis. An answer can use the approved formula while still misunderstanding the user’s business question. Shared calculation logic narrows one source of error, but it cannot decide whether the requested metric fits the decision.
Organizations already organizing technical decisions through a searchable AI knowledge base will recognize the distinction. A formula and the reasoning behind it serve related purposes. Keeping the reasoning accessible helps colleagues interpret changes without reconstructing the original discussion.
For enterprise buyers, the immediate consequence is organizational rather than speculative. A pilot needs participation from business metric owners, reporting specialists, and platform administrators. Evaluating it solely as a connector exercise would leave the most consequential ownership questions unanswered.
Power BI remains a model, even when definitions start elsewhere
The integration preserves a downstream model, so compatibility depends on how faithfully publication represents the upstream definitions.
The publishing requirements describe specific eligibility conditions. Columns used by measures must appear as dimensions, and fields must have names. Missing inputs can cause measures to be dropped. The documentation also identifies unsupported patterns, including multi-source metric views and certain window calculations.
These constraints change what a successful publication means. An operation can produce a usable model without carrying every intended measurement. Teams therefore need to compare the resulting field inventory with the business requirements, rather than treating completion of the operation as proof of completeness.
An illustrative financial report makes the distinction clear. A total might publish successfully while a specialized comparison does not. Users could still open the report and explore available fields. The missing comparison would remain a business problem, even though the integration had performed its supported work.
This is why the shape of the downstream model deserves attention. Report authors work with the fields they receive, and their interpretations depend on the available relationships and calculations. Translation quality must be assessed through the report’s analytical behavior.
Microsoft’s semantic model compatibility guidance supplies relevant competitive context. It warns that layering semantic systems can produce incorrect results when relationship and aggregation behavior conflict. Microsoft also describes Power BI models as resources for external tools and AI consumption through programmable interfaces.
That position complicates any suggestion that customers face a simple choice between an open upstream layer and an isolated reporting application. Both vendors present their models as reusable business infrastructure. Their disagreement concerns where authority belongs and which integration patterns preserve correctness.
The preview should consequently be evaluated as a particular publishing mechanism. It does not establish universal compatibility between semantic products. A supported path for selected definitions says little about an unrelated model arrangement or an existing report with different assumptions.
For customers, the useful comparison is between operating approaches. Keeping definitions primarily within Power BI aligns authorship with existing reporting expertise. Maintaining them upstream aims to serve a broader collection of consumers. The better fit depends on where shared business logic actually needs to travel.
A team serving only established reports might see limited benefit from relocating authorship. A team repeatedly rebuilding the same approved measurements across applications has a clearer reason to investigate. Neither situation establishes a universal winner.
The evaluation should also include exceptions. If a required measurement falls outside the publishing path, the organization needs an explicit handling decision. Quietly rebuilding that measure downstream would reintroduce the split ownership the project was intended to reduce.
Unity Catalog metric views therefore need to earn trust through representative analytical cases. Totals, filtered results, and business-specific comparisons should agree for reasons the team can explain. Correctness is more persuasive when it survives ordinary report interactions.
DirectQuery and Import create different governance boundaries
Shared metric definitions do not create identical access controls across storage modes.
DirectQuery, a mode that sends report queries to the source, keeps query execution connected to Databricks. Import, a mode that stores a data copy in Power BI, creates a separate access boundary. The preview documentation says Unity Catalog policies apply to DirectQuery, while imported data follows Power BI permissions.
That distinction should shape the governance conversation from the beginning. An organization cannot infer that protecting the source automatically protects every downstream copy under the same rules. It needs to identify the system responsible for access at the point where users consume data.
An illustrative regional sales report exposes the issue. A manager should see the authorized regional scope, while another manager should receive a different scope. Testing only with an administrator would provide little evidence that the deployed reporting arrangement respects those intended boundaries.
The relevant question is which identity the connection uses and how that identity relates to the audience. Microsoft’s DirectQuery considerations explain that published reports can use configured shared credentials. DirectQuery by itself does not establish individual-user security. Authentication and model configuration remain part of the outcome.
This does not contradict the value of source governance. It clarifies what source governance evaluates. A source system applies rules to the identity presented to it. Teams still need to establish whether that identity represents an individual viewer, a shared connection, or an approved service account.
Import requires a separate examination because users interact with copied data. Reviewers should ask who can access that copy, which restrictions apply within Power BI, and who maintains them. Those questions remain necessary even when the imported measurements originated from approved definitions.
The choice also affects operational expectations. DirectQuery ties report interactions to source execution, while imported data requires refresh management. Microsoft notes that source performance influences DirectQuery responsiveness. Neither mode deserves a blanket performance conclusion without an actual workload.
For a business team, the practical difference appears during ordinary use. Users expect filters to respond, reports to open predictably, and permissions to reflect their responsibilities. A centralized definition is useful only if the delivery arrangement supports those expectations.
Performance evaluation should therefore include representative interactions and access patterns. An isolated query can reveal calculation behavior, but it does not capture a report containing several dependent visuals. Testing should resemble the work the organization expects people to perform.
Databricks also discusses future performance benefits from materialization, which stores precomputed results for reuse. Its announcement describes Power BI benefiting from those optimizations as forthcoming. Buyers should separate that roadmap statement from capabilities available in their enrolled preview.
This is the central tradeoff in the Databricks Power BI integration. Central authorship reduces duplicated business logic, while execution and storage choices still determine important operational boundaries. A credible pilot makes both sides visible to the people responsible for approving deployment.
Synchronization still needs an operated publishing process
A shared definition remains useful only when downstream consumers receive its approved changes through a dependable process.
Databricks says publication runs from Unity Catalog toward Power BI. Edits made to published measures in Power BI do not flow back upstream. Definition changes require republishing, either manually or through automation. That direction establishes a source of authority, but it does not eliminate update management.
The distinction between current data and current definitions is especially important. A report might query recently updated records while retaining an older published calculation. Conversely, its model might reflect a newly published definition while an imported data copy awaits refresh.
Lakeflow Jobs, Databricks’ workflow scheduler, provides a way to coordinate these operations. The Power BI task documentation describes automated model publication and refresh options. It also distinguishes metadata updates from refreshing imported data, and identifies credentials and task execution permissions as requirements.
For operators, the implication is straightforward: synchronization becomes a maintained workflow. Someone needs responsibility for failed runs, expired credentials, and changes that require attention. Moving publication into a scheduled job makes those responsibilities more visible; it does not remove them.
Imagine an illustrative change to the definition of an active customer. The approved measurement starts excluding accounts under review. Before publishing, the owner needs to consider which reports depend on the field and how users should interpret differences from previous results.
After publishing, the team needs evidence that the intended model received the change. It should also confirm that affected reports still express the approved business question. A successful task run addresses execution; business acceptance addresses whether the resulting report remains appropriate.
One-way publication introduces another practical issue: local experimentation. Analysts often need to investigate alternative calculations before proposing an approved metric. Organizations should distinguish exploratory work from published definitions so users understand which measurements carry organizational approval.
Without that distinction, an attractive downstream edit can become an unofficial replacement. Its results might circulate while the upstream owner remains unaware of the new interpretation. The absence of automatic writeback makes an explicit proposal and approval process more important.
Recovery planning also deserves attention. If a definition produces unexpected results, the team needs to know how it will restore an approved state. That includes the business explanation as well as the model operation. Restoring a calculation without explaining the reporting impact leaves users uncertain.
None of these questions require treating the preview as unsuitable. They provide a realistic account of what centralized authorship entails. The value comes from replacing scattered implementations with a controlled dependency that teams can understand, operate, and review.
The strongest case for Unity Catalog metric views emerges when that process already fits the organization’s working practices. Teams accustomed to reviewed data products have a foundation for managing shared definitions. Teams without clear metric ownership will need to establish it alongside the technical evaluation.
Watch availability, model coverage, and repeatable results
The next evidence should show broader access, better coverage, and dependable behavior in representative reporting workloads.
The first signal is availability. Over the coming months, watch for a documented change from private preview and clearer access conditions. Broader access would strengthen the case for routine evaluation. Continued restricted enrollment would keep adoption dependent on an account-led preview process.
Availability alone would not settle technical suitability. It would, however, make it easier for organizations to compare experiences and assess deployment requirements. Buyers should distinguish an expanded release from a statement of intent, especially when planning work outside an existing preview agreement.
The second signal is coverage of real reporting requirements. Watch the documented limitations and supported patterns, particularly where current restrictions affect existing business measurements. Broader support would strengthen the argument for upstream authorship. Persistent exceptions would increase the amount of special handling customers must maintain.
The useful unit of progress is a complete reporting need. Adding a calculation pattern matters when it removes a specific obstacle from an organization’s model. A longer feature list offers weaker evidence if required reports still depend on unsupported behavior.
The third signal is repeatable operational evidence. Look for evaluations that connect calculation agreement, permissions, publication behavior, and report responsiveness. Results from representative workloads would strengthen the integration’s practical case. Persistent discrepancies or fragile update processes would weaken it, even if initial publication appeared straightforward.
These signals should be evaluated together. Wider availability with narrow compatibility creates a different opportunity from limited access with strong technical coverage. A customer’s decision also depends on whether its reporting estate resembles the workloads being demonstrated.
The Databricks Power BI integration offers a focused proposition: keep approved metric authorship upstream while continuing to serve Power BI users. Its value depends on how much duplicated logic that arrangement removes and how well the remaining operational responsibilities fit the organization.
Start with a report whose metric definitions already cause reconciliation work. Identify the business owner, approved calculation, reporting audience, and expected update behavior before evaluating the publishing path. Compare results across relevant filters, then examine permissions and the handling of a definition change. Include the reporting team in that review so success reflects actual analytical needs. Keep a record of exceptions instead of hiding them in downstream calculations. The useful question is whether shared authorship makes the report easier to trust and maintain. Which existing report would provide the clearest evidence that your organization can manage that responsibility?



