Table of contents
At a glance
When and how do you test a Power BI report for GxP use?
Assess a Power BI report on a risk basis when it informs a GxP decision. Trace source data through transformations and the model to metrics, filters and display. Include data currency, refresh failures and permissions. A plausible chart alone does not demonstrate complete or correctly associated data.
Scope and Applicability
The article covers configured reports and semantic models. An advisory management dashboard may need different evidence from a report used to decide on quality measures.
The correct individual value may appear in the wrong filter context.
Challenge relationships and filters with counterexamples. Unmatched records can disappear and multiple matches can multiply values. Use controlled source data with independently known expected results and compare the relevant views.
Relevance and access are separate controls. Demonstrate the data version pertinent to the statement, not merely the current calendar date. Row-Level Security applies in the Workspace for Viewer roles, but not for Admin, Member, or Contributor roles. Therefore, role checks must reflect the actual, current permissions.
Purpose of report and decision to be defined first
A dashboard providing general oversight has a different impact compared to a report upon which batch prioritization or quality actions are triggered. Describe the specific decision-making process, its input sources, and the potential consequences of using outdated or incorrect data. Accordingly, identify the necessary evidence for model validation, updates, and presentation.
Capture data source, transformation, semantic model, key performance indicators, filters, report, and access pathway. A correctly presented diagram may be based on incomplete data. An accurate data source can be distorted by an incorrect join or filter. Therefore, the evaluation must consider the entire required data chain.
Define the boundaries for import, other used connection modes, and update pathways based on your actual solution. This document does not assume a single, uniform refresh mechanism. Document the gateway and technical identity if they are in use. The visible report alone does not fully describe these dependencies.
Verify metrics against a known reference population
Define each critical number professionally: counters, denominators, period of time, status selection and handling of missing data. "Open deviations" can only include currently open cases or reopened processes. Without this definition, a numerical value cannot be judged as correct.
Prepare a small, known reference dataset. Include cases both within and outside defined time periods, varying statuses, and missing assignments. Calculate expected values independently of the report. The test selection should cover critical transformations and filtering effects, not merely generate a plausible total count.
| Reference case in the example | Expected classification | Validation aspect |
|---|---|---|
| Open case within the period | Included in the numerator | Status and Period |
| Closed case | Excluded by definition | Status filter |
| Reopened case | In accordance with established rules, the procedure is handled | Rules governing process status |
| Case without location | Visible and resolved rather than unnoticed lost | Join and missing assignment |
| Cut-off date | Assigned to defined time zone | Time filter and daily limit |
Example and inspection table: correct data, incorrect denominator
Ten cases fall within the considered time period, two of which are open. The internal metric should indicate 2 divided by 10, i.e., 20 percent. Applying a filter for "open" inadvertently affects both the numerator and denominator. The report shows 100 percent. All visible individual rows are correct; nonetheless, the metric is technically inaccurate. The incorrect denominator arises from the definition of the metric and the filter condition; this example does not describe a Power BI defect and is not a conducted product test.
Record the reference cases and an independent calculation. Test the metric without a filter, with a site filter and with a selection that contains no matching cases. Define how to display a result when the denominator is empty. Displaying zero could hide the fact that a value cannot be calculated. The business definition determines the expected behaviour.
Add a drill-down or an appropriate comparison view if it is relevant to the intended review. The sum of the details should align with the displayed key figure. A visual formatting error can also be significant if percentage signs, units, or rounding affect the decision. The review should explicitly include these elements.
Describe the key figure first without DAX or model terms: What counts in the counter, what is in the denominator, which period of time? In case of a different result, source, transformation, model and representation are examined separately. After a correction, relevant filter and access cases again belong in the comparison.
| Test | Expectation | Possible finding |
|---|---|---|
| Total inventory | 2 out of 10 equals 20 percent | Filter changes the denominator |
| Site selection | Only a defined site scope | Unintended mixing of areas |
| Empty selection | No calculable incidence rate detectable | Zero is reported as the measured value |
| Detailed comparison | Individual steps explain the metric | Rows are lost during the Join operation |
| Export | Significance and filter reference remain identifiable | Release file loses context |
Check update chain and subject-matter data age
An update may include source, export, gateway, semantic model, and report level. Define which time point supports which assertion. A model refresh at 08:00 does not prove that the source data also originates from 08:00. Investigate a controlled case involving outdated source data and technically successful data transfer.
Microsoft describes gateway connectivity, login credentials, and the schedule as components of the planned update. The schedule is disabled after four consecutive failures or a non-resolvable configuration error. Additionally, the planned update is paused after two months if no report or dashboard is invoked; the model owner receives an email notification in this case. Verify error detection and controlled resumption, including for infrequently used reports. The professionally permitted data age limit aligns with your process, not a general manufacturer time window.
In addition, evaluate partial updates and missing data segments, to the extent that the architecture supports such scenarios. The dashboard must not present an incomplete or outdated inventory as a current decision-making basis without proper indication. Appropriate controls may include visible source snapshots, monitored errors, and regulated usage limits. The allowed recency and frequency are derived from your process.
Own failure example: The report is opened, the data are missing
In this example, current lock status is required for a daily decision. The most recent validated source originates from the previous day. The report opens seamlessly and displays the outdated population. The test simulates controlled source failure and verifies that the user can detect the reduced data state and apply the predefined fallback decision. A green report loading action would not constitute an appropriate success criterion. The outdated state arises due to source unavailability; this example does not describe a Power BI defect and is not a completed product test.
The Microsoft documentation also differentiates between notification pathways. On-demand and programmatically triggered refreshes generate different email error notifications than the planned updates described. Therefore, verify the actual alerting behavior of your trigger pathway, including recipients and designated representatives. Sending a notification to a deprecated owner account constitutes no valid operational response.
| Controlled Error | Expected finding | Evidence |
|---|---|---|
| Source unchanged, model refresh successfully completed | Data age remains identifiable | Source and reporting responsibility |
| Gateway not reachable | Updates and usage are appropriately managed. | Error status, recipient, and response |
| Schedule deactivated | No unnoticed permanent operation with old status | Detection and controlled reactivation |
| Scheduling paused after two-month period of inactivity | Pauses are also detected in infrequently used reports. | Notification of the model owner and controlled reprocessing |
| Programmatic refresh fails | Appropriate monitoring pathway is applied | Actual triggered alarm |
| A data source is missing | The required population is not clearly and fully specified. | Source-based completeness comparison |
Sources: [2]
Test actual roles with row-level security
In a workspace, row-level security restricts data for users with Viewer access. It does not filter rows for Admin, Member or Contributor roles. Test real Viewer accounts and document which editing roles have wider access. A successful “View as” test does not cover every real access path.
Review positive and negative data areas. A user at Site A shall have access to required data but shall not be able to view unauthorized data from Site B. Consider multiple role assignments, authorizations, and exports, to the extent they are part of usage. The specific combination and platform configuration determine the scope of access.
Distinguish information protection from professional release. RLS is not an electronic signature and does not constitute confirmation of a quality decision. A duly authorized user may view a flawed report. Therefore, the suitability of the metric and the appropriateness of data access each require independent validation evidence.
Monitor and manage model changes and transfers
A new column may impact relationships, filters, or existing key performance indicators. Jointly evaluate changes to data source, transformation, model, and report. The current state should be traceable to the reviewed reference cases. An unchanged diagram interface does not automatically imply unchanged calculation logic.
Provide key performance indicator definitions, data sources, update rules, roles, and known limitations. An internal reviewer should be able to trace an outlier value back to its associated data. A visually appealing visualization lacking a clear rationale or data date may generate additional questions during an audit.
Obtain required evidence independently of mere screen states. Depending on the intended use, configuration state, reference data, and test results may be required. The exact retention period follows your process. A regularly overwritten dashboard does not automatically constitute complete historical documentation of a prior decision.
Next steps
- Define the decision, metric and data scope.
- Capture transformations, relationships, and calculations transparently.
- Test filters, error handling and user roles against a known dataset.
- Update, version release, and changes control.
Project experience and collaboration
Recalculate the metric independently and test access with actual roles
With your process owners, we select a known dataset and define what a critical metric means. We compare the calculation, filters and display against an independently established expectation. Depending on intended use, we add cases for stale or missing source data and test actual viewer permissions. With IT, we follow a failed refresh and its recovery. This shows whether the report supports the required decision using the correct data version.
Frequently asked questions about Power BI validation and data refresh
Is the comparison with the previous report sufficient?
Must every Power BI dashboard be validated?
The evaluation is based on the intended use. A GxP-relevant report that supports quality decisions requires appropriate evidence to substantiate its conclusions and usage. A general-purpose information dashboard may be appropriately categorized differently. Document the purpose, data, and decision, rather than inferring the classification solely from the product name.
What does a successful refresh demonstrate regarding data quality and currency?
He confirms the technical update step, but does not automatically ensure a current, complete source inventory or correct transformation and key performance indicators. Verify source data, relationships, filters, and the chain of custody down to the display level. Even a successful import may inherit outdated data. Subject-matter age limits and handling of incomplete sources must align with the supporting decision-making process.
Why should a Power BI test include empty data sets?
Since an empty denominator or missing data region implies a different interpretation than a measured zero value, verify the intended data representation and its clarity. An attractive diagram must not silently treat missing foundation as a benign result.
Must exports from Power BI be individually verified?
To the extent that they are part of the intended use, consider their content and context. Filters, units, data status, or explanatory information may appear differently in a given output. A correct interactive report does not automatically validate every exported file. Selection is based on the actual decision supported.
Does each failed refresh generate the same email alert?
No. Microsoft differentiates between trigger mechanisms; on-demand and programmatically initiated refreshes do not generate the described corresponding email error notifications. Verify the actually used monitoring pathway. Recipient, representative, detection of a disabled schedule, and controlled resumption are part of the operational justification.
Sources: [2]
Sources and review date
Microsoft documentation explains product functions and their limitations. Check updates, roles and notifications on the actual model and service used. The numerical examples, acceptance criteria and tables are their own implementation aids, not standard rules.
- European Commission: EU GMP Annex 11, Computerised Systems (2011)Current version as per EudraLex inventory as of 14.10.2026. Application, risks, operation, and data lifecycle; specific testing proposals are derived independently.
- Microsoft: Data refresh in Power BIUpdate pathways and differences in their monitoring. For on-demand and program-triggered updates, the described email error notifications do not apply; verify the actual monitoring pathway.
- Microsoft: Row-level security in Power BISupplier source for product functionalities. Availability and behavior depend on version, edition, license, and configuration; no general compliance certification.
- Microsoft: Configure scheduled refreshGateway, login credentials, refresh history, and deactivation upon four consecutive failed attempts or uncorrectable configuration. No SLA commitment.
Sources checked: .
Check key performance indicator, data status, and access rights together
With a critical key metric, its professional definition, and the current update pathway, we assess reference cases, data age, and actual access roles.
- Cloud and SaaS validation →
Maintain the validated state of cloud and SaaS systems as suppliers issue new versions. We define the boundary between your responsibilities and the supplier’s and agree how each release will be assessed.

