Is your data truly ready for Copilot? According to Greg Trzeciak of Pragmatic Works, the answer might surprise you. In this training session, Greg walks through one of the most overlooked but powerful ways to prepare Power BI data models for AI usage: synonyms. As organizations increasingly rely on Copilot and natural language interactions with data, proper preparation becomes essential for accurate, trustworthy insights.
Greg begins by addressing a common misconception. When Copilot returns confusing or incorrect results, the issue is rarely the AI itself. Instead, the root cause is often unprepared data models. As AI becomes a primary interface for analytics, the role of the data analyst continues to evolve. Preparing data for AI consumption is no longer optional. It is foundational.
Unprepared data can lead to several challenges, including:
Greg emphasizes that Copilot should be thought of as a capable intern. It understands Power BI, modeling concepts, and analytics patterns, but it does not understand the unique language of your organization unless you teach it.
Synonyms are exactly what they sound like. They are alternative words or phrases that Copilot can use to interpret natural language questions. In Power BI, synonyms can be applied to tables, columns, measures, and fields. This allows Copilot to map different terms used by business users to the correct elements in the data model.
Greg explains that organizations often use internal language, nicknames, acronyms, or informal terms that do not exist in the data model. Without synonyms, Copilot has no way to connect these terms to the correct data.
Examples of how synonyms help include:
Although the Power BI Q&A visual is being deprecated in December 2026, its functionality remains critical. Greg highlights that the Q&A setup experience directly powers Copilot’s ability to understand natural language queries. By configuring synonyms within Q&A, users are effectively training Copilot.
The synonym pane in Power BI provides visibility into all tables, columns, and measures. It also offers suggested synonyms, which can be approved or customized. Analysts can easily add organization-specific terms to ensure Copilot responds accurately.
Greg demonstrates the impact of synonyms using a sales data model. When asking Copilot, “What are my total sales by reseller?”, Copilot correctly returns a clean and accurate visualization. However, when the same question is asked using the word “friends” instead of “reseller”, Copilot fails because the term does not exist in the model.
To resolve this, Greg navigates to the Q&A setup in model view and adds “friends” as a synonym for the reseller name field. He carefully avoids ambiguity by ensuring synonyms are applied to the most appropriate column.
Once configured, Copilot is able to successfully answer the same question using the term “friends”. The data, calculations, and results remain unchanged. Only the language interpretation improves.
Greg shares several practical tips for getting the most value from synonyms:
By following these practices, organizations can significantly improve Copilot’s reliability and user adoption.
Greg concludes by reinforcing that synonyms are a small effort with a massive impact. They bridge the gap between business language and technical models, making AI-driven analytics more intuitive and trustworthy. As chat with your data becomes the norm, thoughtful preparation ensures Copilot delivers meaningful, accurate insights.
This session is part of Pragmatic Works’ ongoing “Chat With Your Data” series, designed to help teams unlock the full potential of AI and Microsoft Fabric. With the right preparation, Copilot becomes not just a tool, but a powerful analytics partner.