For all the advances we've made in business intelligence (BI) over the years, the way users consume analytics really hasn't changed all that much. IT teams still spend considerable time building reports and dashboards, defining KPIs, and providing tools to monitor business performance. Much of this effort focuses on tracking lagging indicators, helping users understand what happened last week, last month, or last quarter.

While knowing what happened is important, it's usually just the starting point. Once we identify an unexpected trend or a change in performance, the natural next step is to understand why it happened, what might happen next, and what we can do about it. That's where traditional reporting experiences can start to feel restrictive. Even with interactive filters and drill-down capabilities, we're largely confined to the questions and analytical paths someone anticipated when the report was designed.

Conversational analytics is starting to change that dynamic. Advances in generative AI, semantic modeling, and technologies like Microsoft Fabric IQ are making it possible to interact with business data in more natural ways. Rather than navigating through predefined reports, users can ask questions, explore follow-up questions, and investigate relationships across their business data as their curiosity takes them.

In this article, we'll explore how conversational analytics is reshaping BI, how existing investments in tools like Power BI can be leveraged to create these new experiences, and why giving AI the right business context is essential to making it all work.

One Question Leads to Another

One of the challenges with traditional BI is that we don't always know what questions we're going to ask until we start looking at the data. A sales report might show that revenue is declining in a particular region, but that observation naturally raises other questions. Is the decline concentrated among a handful of customers? Are certain product lines underperforming? Is this part of a broader trend, or something unique to that region?

With traditional reporting tools, answering those follow-up questions depends largely on how the underlying reports and dashboards were designed. If the right filters, measures, and drill-down paths are available, users may be able to find what they're looking for. Otherwise, they're left exporting data to Excel, building their own analysis, or asking the BI team for help.

Conversational analytics offers a different approach. Rather than navigating through predefined reporting experiences, users can ask questions in natural language and refine their analysis as they go. Each answer provides an opportunity to explore another angle, investigate an unexpected result, or follow a new line of inquiry without having to start over.

What's particularly interesting is how this changes the relationship between business users and their data. Instead of relying entirely on someone else to anticipate their information needs, users can take a more active role in exploring the business questions that matter to them. And as these capabilities become more deeply integrated into the applications people already use, those conversations won't necessarily have to start in a traditional BI tool.

The Intelligence Behind the Dashboards

The good news about this shift toward conversational analytics is that you don't necessarily have to start from scratch. In fact, you probably already have many of the foundational components needed to make these experiences possible.

Behind most enterprise BI solutions is a semantic model that gives structure and meaning to the underlying data. These models define how different datasets relate to one another, how business metrics are calculated, and how information should be organized and interpreted. For example, a Power BI semantic model might define how revenue is calculated, how customers are grouped into regions, or how sales performance is measured across different product lines.

Historically, analysts have tended to build these models around the needs of specific reports and dashboards, without giving much thought to how they might be used elsewhere. But the business knowledge captured within them has value well beyond traditional reporting. With AI-powered analytical experiences, semantic models can help provide the definitions, relationships, and calculations LLMs need to interpret natural-language questions and produce answers that align with how the business actually measures performance.

This creates an interesting opportunity for organizations with established BI environments. Rather than rebuilding years of business logic for a new generation of AI tools, they can begin extending their existing investments into conversational analytics. Technologies like Microsoft Fabric and Power BI are making it possible to reuse these models in new ways, providing a foundation for more intelligent data agents and conversational experiences.

Figure 1: Example of a Semantic Model in Microsoft Fabric

Of course, semantic models are only part of the equation. While they do a good job of defining how data should be analyzed, they don't necessarily capture the broader business context needed to understand how everything fits together. That's where things start to get particularly interesting.

Context Changes Everything: Teaching AI How Your Business Works

As valuable as semantic models are, they don't necessarily tell the whole story. A model might know how to calculate revenue by customer, compare sales across regions, or measure profitability by product line. But understanding how those numbers relate to what's happening across the business is another matter entirely.

Consider a manufacturing company that's experiencing delays in fulfilling customer orders. A traditional BI model might help identify which orders are running behind schedule or how much revenue is at risk. But what if we want to understand which customers are most likely to be affected by a supplier disruption? Answering that question requires a much broader understanding of how customers, orders, products, materials, suppliers, and manufacturing operations relate to one another.

Making matters worse, most ERP and line of business systems rely on highly complex and normalized database schemas, with business information spread across dozens or even hundreds of related tables. Understanding how those tables fit together, which relationships matter, and how to interpret the underlying data often requires specialized knowledge of the application itself. Anyone who's tried to build a report directly against an ERP database knows how quickly things can get complicated.

And while modern LLMs are remarkably good at generating queries and navigating unfamiliar data structures, they don't necessarily understand the business logic buried within these systems. Giving an AI agent access to an ERP database doesn't mean it suddenly knows how to trace a customer order through production, procurement, inventory, and fulfillment.

This is where technologies like Microsoft Fabric IQ start to get interesting. Fabric IQ introduces capabilities for modeling business entities and their relationships using what's known as an ontology. Rather than expecting an AI agent to figure out how dozens of underlying tables fit together, an ontology provides a more meaningful representation of the business itself. Knowledge graphs can then help connect those entities, allowing agents to navigate relationships and investigate questions that span multiple business processes.

Figure 2: Navigating Complex Relationships Using Graph in Microsoft Fabric

For example, an AI agent equipped with the right business context could potentially trace a delayed shipment back through the affected production orders, identify a common supplier dependency, and determine which customer commitments might be at risk. Instead of trying to piece together the underlying ERP data model on its own, the agent can follow established business relationships to guide its investigation.

Of course, this context doesn't appear automatically. Someone still needs to define and validate the relationships, establish consistent business definitions, and make sure the underlying data can be trusted. But as these capabilities mature, they offer an interesting path toward AI experiences that can do more than answer questions about numbers. They can begin to understand how those numbers relate to the way your business actually operates.

Meet Your New Data Analyst

With the right data and business context in place, the next question is how we make all this intelligence accessible to the people who need it. This is where AI-powered data agents come into the picture.

Think of a data agent as a virtual analyst that can help users explore business information through natural conversation. Rather than opening a report, applying filters, or trying to figure out which dashboard contains the information they need, users can simply ask a question. The agent interprets the request, works with the data sources and business context available to it, and returns an answer that users can explore further with follow-up questions.

Figure 3: Data Agents in Microsoft Fabric

Microsoft is bringing these capabilities together through Fabric data agents and integrations with experiences like Microsoft 365 Copilot and Copilot Studio. This creates opportunities to make conversational analytics available in the applications people already use, rather than requiring everyone to become proficient with Power BI or other specialized analytical tools.

For example, consider the finance scenario illustrated in the Microsoft demo below. A finance leader can use Copilot to explore financial performance, investigate budget variances, and ask follow-up questions to better understand the factors driving those results. With access to trusted financial data and the right business context, an agent can help connect the dots and guide the investigation without requiring the user to navigate multiple reports or applications.

What's particularly interesting is how this could change the role of traditional BI teams. Rather than spending so much time responding to requests for new reports, additional filters, or slightly different views of existing information, analysts can focus more of their attention on maintaining trusted data models, defining business metrics, and making information accessible through reusable analytical services. These concepts are illustrated in a real world use case in the video below.

From Insight to Action

As conversational analytics continues to evolve, one of the more interesting opportunities is the ability to move beyond simply answering questions and start doing something with the answers.

Consider the finance scenario from the previous section. After investigating an unexpected increase in operating expenses, a finance leader might ask an agent to identify opportunities to reduce costs, evaluate the potential impact of different scenarios, or prepare a revised forecast. Rather than stopping at the analysis, the conversation could naturally progress toward recommendations and next steps.

And with the right integrations in place, there's an opportunity to take things even further. AI agents can potentially initiate workflows, create tasks, generate reports, or interact with enterprise applications to carry out approved actions. For example, an agent might prepare a budget adjustment for review, notify department managers about spending exceptions, or initiate an approval process without requiring the user to switch between multiple systems.

Of course, moving from analysis to execution introduces a different set of considerations. It's one thing for an AI agent to answer a question incorrectly. It's another for that agent to make changes to financial records, update business transactions, or trigger operational processes based on incomplete or misunderstood information. This is something we've been working extensively with clients on lately, particularly around establishing the right guardrails for agent-driven processes, including permissions, approval workflows, auditability, and human oversight.

Still, the direction is compelling. As conversational analytics becomes more deeply connected to the systems and processes that run the business, the line between analyzing information and acting on it begins to blur. The opportunity is to make business intelligence a more active part of how work gets done, rather than something we consult periodically to see how things are going.

Getting Your Data Ready for the Conversation

As promising as conversational analytics and AI-powered agents are, their effectiveness still depends on the quality of the information and business context available to them. Giving an agent access to more data doesn't necessarily make it more intelligent, especially when that data is incomplete, inconsistent, or poorly understood.

If you're wondering how to start taking advantage of these capabilities, I'd suggest taking stock of the data assets you already have. Existing Power BI semantic models, established business metrics, and well-maintained data pipelines can provide a useful foundation. The challenge is making sure those assets are trustworthy, accessible, and organized in ways that support a broader range of analytical experiences.

Data governance becomes particularly important as we move away from predefined reports and dashboards. Traditional BI environments often limit what users can see through carefully designed reporting experiences. Conversational agents introduce more flexibility in how users explore information, which means we need to pay even closer attention to access controls, data sensitivity, and how security policies are enforced across different systems.

There's also the question of consistency. If sales, finance, and operations each have different definitions of revenue, profitability, or customer performance, an AI agent isn't going to resolve those disagreements on its own. Establishing shared business definitions and maintaining the semantic models and ontologies that represent them will become increasingly important as more users begin interacting with enterprise data through AI.

Fortunately, you don't need to solve every data management challenge before getting started. A practical approach is to begin with a well-understood business domain, build on existing data and semantic models, and expand as the organization gains confidence in the results. The lessons learned along the way can help guide broader investments in data quality, governance, and business context.

The technology is advancing quickly, but the underlying work of making enterprise data reliable and understandable remains just as important as ever.

Closing Thoughts

We've spent years building increasingly sophisticated reporting environments, yet getting answers to relatively straightforward business questions can still be surprisingly difficult. The information is often there, but finding it, connecting it, and making sense of it can require more effort than it should. Conversational analytics offers an opportunity to make that process feel a lot more natural.

What's particularly interesting is how these capabilities are evolving beyond simply asking questions in natural language. With technologies like Microsoft Fabric IQ, semantic models, and AI-powered agents, we're beginning to give these systems a much richer understanding of how our businesses actually work. That opens the door to more meaningful analysis, better-informed decisions, and eventually the ability to take action without having to jump between multiple applications.

Of course, there's still work to be done around data quality, governance, and making sure AI-generated answers can be trusted. However, if you’ve already invested in building a solid BI foundation, you may be further along than you realize. Much of the business knowledge needed to support these experiences may already exist within your existing data models and reporting environments.

In the months ahead, we’re going to continue to see a huge shift towards reimagined BI experiences which mix-and-match traditional analytics with conversational analytics, transaction processing, and agentic automation. From an end user perspective, it’s very exciting to imagine a world where BI tools have a better understanding of the people that are using them.