4.3.2. AI Visuals and Copilot Model Summaries
💡 First Principle: AI visuals apply statistical models inside the report so readers can ask "what drives this?" and "how does this break down?" interactively. Copilot adds a conversational layer that can explain the model and summarize data in plain language. Both are aids to judgment, not replacements for it.
| AI feature | Question it answers | How it works |
|---|---|---|
| Key influencers | What factors affect this outcome, and what segments have the highest values? | Ranks fields associated with a metric increasing/decreasing or a categorical outcome; Top segments tab shows combinations with the highest outcome |
| Decomposition tree | How does this total break down across dimensions? | Interactive tree; users expand by any dimension, or use AI splits (High value / Low value) to let Power BI choose the next dimension |
| Q&A visual | Natural-language question to a visual | Legacy natural-language feature; Microsoft is retiring Q&A in February 2027 (extended from December 2026) in favor of Copilot |
| Anomaly detection | Which points are unexpected? | Line chart Analytics pane (4.3.3) |
| Narrative with Copilot | What's the story on this page? | Copilot-generated summary (4.1.4) |
| Copilot pane | Questions about the model or data | Conversational answers and summaries |
Key influencers analyzes an Analyze field (a measure, or a categorical outcome like Churn = Yes) against Explain by fields. The Key influencers tab shows, for example, "When Tenure is less than 6 months, the likelihood of churn increases by 2.3x." The Top segments tab clusters combinations of conditions. Results reflect association within the data you supply.
Decomposition tree is ideal for root-cause exploration and ad hoc slicing: start with a measure, then expand by Region, then by Product, and so on. AI splits pick the dimension with the highest or lowest value next; lock levels to fix the analysis path for readers.
Use Copilot to summarize the underlying semantic model: in the Copilot pane (in Desktop or the service, on a supported capacity), ask Copilot to describe the semantic model, for example "Summarize the semantic model" or "What tables and measures does this model contain?" Copilot returns an overview of the data, its tables, key measures, and what questions it can answer, which helps report authors and consumers orient themselves in an unfamiliar model. Copilot can also answer questions about data in the report and create summaries. Like all Copilot features, it needs paid capacity (F2/P1 or higher, directly or through a Fabric Copilot capacity), tenant enablement, and benefits from good names, descriptions, and synonyms.
⚠️ Exam Trap: "Allow users to explore a measure by any dimension in any order, with Power BI suggesting the next split" is the decomposition tree (AI splits), while "identify which factors most affect customer churn" is key influencers. Choosing the wrong AI visual is a common distractor pattern.
Reflection Question: Key influencers reports that customers who contact support more than three times are 2x more likely to churn. Why is it a mistake to conclude that reducing support contacts will reduce churn?
