Microsoft's Build 2025 conference brought two Power BI Copilot capabilities that represent a meaningful shift in how analytics is surfaced to end users — not incremental feature additions, but structural changes to the interface model. The standalone Power BI Copilot Chat with Your Data experience makes conversational analytics the primary entry point for data exploration, available to any licensed user across all their permitted datasets. The extension of Copilot to securely embedded Power BI reports brings the same natural language interface to organisations that deliver analytics through custom portals and applications rather than the Power BI Service directly. Both features are now generally available — and both require deliberate governance preparation before enterprise rollout.

What Changed at Microsoft Build 2025

Power BI Copilot has existed in various forms since 2024, primarily as a report-authoring assistant and a pane within individual reports that could answer questions about that specific report's data. Build 2025 expanded the scope in two important directions simultaneously.

The first expansion removed the report-level boundary from the Copilot conversation. Previously, a user asking Copilot a question could only draw on the data in the report they had open. The new standalone experience allows Copilot to search across all datasets and reports the user has access to, answer questions that span multiple data sources, and present results without the user needing to know which report contains the relevant data. This is a qualitatively different capability — it moves Copilot from a report companion to a cross-estate analytical assistant.

The second expansion moved Copilot outside the Power BI Service entirely. Organisations that embed Power BI reports in their own web applications, customer portals, or internal tools — using Power BI Embedded — can now surface the Copilot pane within those embedded experiences, providing natural language interaction to audiences who may never access the Power BI Service directly.

"Chat with Your Data is not a dashboard feature — it is a replacement for the dashboard as the primary interface for many users. The question is not whether to enable it, but how to govern the data it accesses."

Chat with Your Data: What It Actually Does

The standalone Power BI Copilot Chat with Your Data experience is a full-screen conversational interface accessible from the Power BI Service navigation. Users type natural language questions — "What were our top five revenue channels last quarter?", "Show me headcount by department as of last month", "Which product categories have declining margins over the past six months?" — and Copilot searches across all semantic models the user has access to, identifies the most relevant data, generates a DAX query against the appropriate model, executes it, and returns a visualisation or narrative response.

The experience is genuinely conversational — follow-up questions refine the previous response, Copilot can explain why it chose a particular metric or time period, and users can ask it to modify a chart type or add a comparison. For business users who know what question they need answered but do not know which report to open or which filter to apply to find the answer, this experience reduces the time from question to insight to seconds rather than minutes.

What Copilot Can and Cannot Do in This Context

Understanding the boundaries of the experience prevents unrealistic expectations from propagating through the business after rollout. Copilot in this context can only access data within certified or shared semantic models that the user has permission to view — it does not access source systems directly, does not read from unmodelled data files, and does not combine data from models the user cannot access. The quality of Copilot's responses is directly proportionate to the quality of the underlying semantic models: well-named measures, clear table descriptions, and properly defined hierarchies produce more accurate and useful Copilot responses than unnamed base measures and undescribed fact tables. This is the data governance argument for semantic model quality investment that organisations should make internally when building the business case for Copilot rollout.

Enabling Chat with Your Data: Admin Steps and Prerequisites

The standalone Copilot experience is off by default at the tenant level and must be explicitly enabled by a Power BI administrator. The enabling path is: Admin Portal → Tenant Settings → Copilot and Azure OpenAI Service → Users can access a standalone, cross-item Power BI Copilot experience. The setting can be applied to the entire organisation or scoped to specific security groups, which allows a phased rollout enabling the experience for a defined pilot group before opening it to all users.

Two prerequisites must be in place before the setting has any effect. First, the organisation must have Power BI Copilot enabled at the tenant level — the parent setting that gates all Copilot functionality within the organisation's Power BI environment. Second, the workspace must be backed by Power BI Premium capacity or Microsoft Fabric capacity — Copilot features do not operate in shared Pro capacity workspaces. For organisations that are still operating primarily on Pro licenses without Premium or Fabric capacity, this is a gating requirement that needs to be addressed before Copilot rollout is possible.

Governance Implications of the Standalone Copilot Experience

The cross-dataset scope of the standalone Copilot experience introduces a governance consideration that report-level Copilot does not: a user who has view access to multiple semantic models can potentially ask Copilot questions that cross-reference data from models they may not have consciously combined before. If a user has access to both the HR headcount model and the financial cost model, they can ask Copilot questions that join headcount data with cost data in ways that produce insights the organisation may not have intended to make readily accessible.

This is not a security flaw — Copilot strictly respects the access permissions defined in each semantic model, and Row-Level Security filters are applied before any Copilot response is generated. But it does raise the question of whether the access permissions currently in place across the organisation's semantic models are correctly configured for an environment where conversational cross-model queries are easy rather than requiring deliberate technical effort. Before enabling the standalone Copilot experience, enterprise administrators should review semantic model access permissions with the assumption that users will explore data more broadly and more freely than they have done with traditional report-based navigation.

Copilot in Embedded Power BI Reports: What Changes

Organisations that deliver Power BI analytics through embedded applications — whether internal tools, customer-facing portals, or partner dashboards — can now enable the Copilot pane within those embedded experiences. When enabled, users interacting with the embedded report see the Copilot pane alongside the report visuals and can ask natural language questions that Copilot answers in the context of that specific report's semantic model.

This matters most for organisations whose embedded analytics audience is broad and varied — executives, customers, or operational staff who are familiar with the organisation's data but not with Power BI's interactive filtering capabilities. For this audience, natural language interaction dramatically lowers the barrier to extracting answers from the data, without requiring the organisation to build separate Q&A interfaces or pre-anticipate every possible user question in the report design.

The embedded Copilot experience is scoped to the data visible in the embedded report — it does not cross report boundaries or access other semantic models in the tenant, which makes its governance implications narrower and more tractable than the standalone cross-dataset experience.

Enabling Embedded Copilot: Configuration and Requirements

Enabling Copilot in an embedded Power BI report requires three conditions to be met simultaneously. First, the organisation's tenant-level Copilot setting must be enabled. Second, the workspace hosting the semantic model must be backed by Power BI Premium or Fabric capacity. Third, Copilot must be explicitly enabled in the embedded report configuration — the property is exposed in the Power BI Embedded SDK configuration object as a Copilot settings flag, and defaults to disabled to avoid surfacing Copilot to embedded audiences without the host organisation's explicit opt-in.

For organisations using the Power BI Embedded SDK to build their embedded analytics experience, the activation is a configuration change in the embed token generation and the report embed settings — no visual redesign or report rebuild is required. The Copilot pane renders as a panel alongside the existing report visuals, and its appearance can be controlled through the SDK settings to match the embedding application's design language.

Data Security Across Both Experiences

Both the standalone Chat with Your Data experience and the Embedded Copilot experience operate within the same security boundary as the underlying Power BI platform. Row-Level Security defined in the semantic model is applied before Copilot generates any response — a user with an RLS filter that limits them to European region data will receive Copilot responses derived only from European region data, even if they ask a question that would logically span all regions. Object-level security, which restricts access to specific tables or columns within a semantic model, is also respected.

Copilot responses are generated using Azure OpenAI Service infrastructure within Microsoft's trust boundary — the data submitted to generate a response does not leave the Microsoft cloud environment and is not used to train the underlying language model. For organisations subject to data residency requirements, it is important to verify that the Power BI and Azure OpenAI Service regions are aligned with any applicable data sovereignty commitments before enabling Copilot for users whose data falls under those requirements.

Enterprise Readiness Checklist

What Enterprise Teams Should Do Before Enabling Either Feature

The technical prerequisites for enabling Power BI Copilot Chat with Your Data and Embedded Copilot are manageable for most organisations already on Premium or Fabric capacity. The more important preparatory work is governance-focused: reviewing semantic model access permissions, validating the quality of measure names and descriptions, and confirming data residency alignment for any models containing regulated data.

For the standalone Chat with Your Data experience specifically, a phased rollout — enabling the feature first for a defined group of power users who can identify edge cases in Copilot's cross-model responses and flag any unexpected data exposures before broader rollout — is the approach that balances the speed of adoption with the governance rigour that enterprise data stewardship requires. The security group scoping available in the tenant setting makes this phased approach straightforward to implement without requiring separate tenant configurations.

If your organisation is assessing readiness for Copilot rollout, needs help with the semantic model quality improvements that make Copilot effective, or wants to design the governance framework for cross-dataset AI analytics access, speak with a certified Power BI consultant at Numlytics. We work with enterprise analytics teams across the US, UK, Australia, and UAE to prepare Power BI estates for Copilot deployment in a way that delivers the productivity benefits without creating unintended data access patterns. For the broader context on AI capabilities in Power BI, see our post on moving from data analytics to AI in Power BI, and for the governance infrastructure that supports Copilot deployment, explore the Numlytics Power BI Governance Platform.