Many new Microsoft Fabric features were recently announced at the second annual Microsoft Fabric Community Conference (FabCon 2025). One of these announcements is particularly impactful for all Microsoft Fabric users, as Copilot in Fabric is being made available to all paying Fabric customers. Previously, this AI-assistant was only available to customers working in a workspace aligned to an F64 SKU capacity, a costly barrier to many would-be adopters.
The democratization highlights the need for Copilot** users to understand the risks and benefits of using the tool and emphasizes the importance of knowing how to mature a Copilot implementation.
In an immature environment, users may observe that this Copilot returns less-relevant results than anticipated, or that its results are unreliable. These concerns arise because, like many tools, Copilot is limited by the environment in which it is deployed, and improving it requires maturing its operating context. As with any AI tool, bad data can be worse than no data, and a bad answer informing business decisions can be worse than no answer.
Data governance: Why Copilot in Fabric is like a new employee
Imagine hiring a new employee who joins remotely without any prior knowledge of the company’s systems or processes. Their onboarding experience is dictated only by documents within a repository and the systems to which they are given access. Their success heavily depends on the quality and accessibility of training materials, which may provide confident answers using old data, or not understand a column, not know that there’s comingled test or legacy data, etc. They may a highly capable employee but provide confidently incorrect deliverables due to stale documentation, organizational acronyms to be learned, things that are figured out over time, other undocumented institutional knowledge, etc. Consider this Copilot to be like that new employee, and its success is dictated by the quality of the organization’s data governance.
If the organization has excellent data governance, the new employee will quickly become productive, accessing accurate datasets and making informed decisions confidently. Conversely, poor governance could result in misinterpreted acronyms, incorrect dataset analyses or confident yet erroneous conclusions drawn from outdated references.
Effective governance allows Copilot to be a powerful enabler. Data governance practices such as curating datasets, providing ongoing training and education, establishing operational guardrails and monitoring and evaluating the tool’s cost, use and performance are all important practices to ensure that Copilot performs effectively. Microsoft Purview provides a seamless integration into the Microsoft Fabric platform to assist with governance, protection, and risk management of enterprise data and now has a direct integration with Copilot. Additionally, Purview can provide further insight into how Copilot accesses and uses data.
Data access and quality
To mature the use of Copilot in Fabric, we believe a multi-faceted strategy is essential. “Garbage in, garbage out” is the evergreen motto for working with AI models and the same is true for Copilot. For example, if a Copilot user is working with a poorly constructed Power BI dataset, their report building experience will be unproductive. Clearly defined and named tables, relationships, measures, etc. provide improved outputs. Similarly, any users generating queries with Copilot against lakehouses, warehouses or other Fabric data stores will benefit from clean and well-maintained data. In most instances when Copilot in Fabric is used to build queries or develop reports, ensuring users only have access to curated, well-maintained datasets and/or data stores will reduce the likelihood of the tool generating results with incorrect or stale data.
In addition to synergizing with the maturation of an organization’s data governance and data management practices, there are several other key accelerators organizations can adopt to efficiently mature Copilot’s use and effectiveness.
Active Copilot in Fabric training
Adoption begins with understanding. Providing active training sessions for employees ensures they are not only aware of Copilot’s capabilities but also the potential risks to mitigate. A well-maintained environment still requires users be trained, but less time is needed if users are only able to use the solution against well-maintained data. Ultimately, Copilot will help users build their data products faster. Training will help ensure that speed is met with quality.
Operational guardrails to manage change
Fundamentally, Copilot will help developers and business users build data products (data pipelines, queries, datasets, reports, etc.) faster. However, data products developed with Copilot need the same guardrails as any other data product. Many organizations struggle with maintaining, testing and deploying code across multiple environments, especially when it comes to data products. Anything moving into production should be reviewed by at least one other qualified person, and data products built with the support of Copilot in Fabric are no different, regardless of their intended audience. Implementing a Center of Excellence can help guide other change management best practices and encourage user engagement and feedback loops to refine adoption and maturation of Copilot.
It is also important to adhere to the access management practices of least privilege. If users have excessive access rights, they may be able to prompt Copilot against datasets they should not be able to access or query data without truly understanding the outputs. These risks can be mitigated by limiting access to only promoted and endorsed content within the Fabric environment.
Monitoring cost, usage and adoption
With the recent Copilot announcement, it is important to ensure all Fabric capacities have sufficient compute available to support its use and avoid capacity overages and increased cost. Monitoring how employees use Copilot can highlight areas for increased adoption and identify best practices. Regularly reviewing analytics on usage patterns enables decision-makers to implement strategies that encourage wider utilization while providing another governance avenue.
Other ways to improve Copilot in Fabric capabilities include:
- Integrating domain-specific data sources that enhance the tool’s contextual understanding
- Developing internal prompts for specific types of tasks
- Developing a team focused on monitoring improving Copilot effectiveness across the organization
Maturing Copilot in Fabric usage involves a comprehensive strategy that requires wide involvement from stakeholders and is dependent on the inclusion of data governance, user training, change management and broader operational oversight. Following these guidelines can transform Copilot into a valuable asset to enhance productivity and data-driven decisioning.
**While there are many Copilots available, in this blog, Copilot refers only to the Copilot in Fabric tool.
To learn more about our Microsoft consulting services, contact us.
