Hello! Welcome back to our workshop preparation on AI-Powered Analytics in Power BI.
In the last lesson, we focused on setting up the Decomposition Tree. You learned how to configure the Analyze and Explain by fields and how to begin an investigation using both manual exploration and AI-assisted splits. This prepared you to teach managers how to start their analysis.
Today, we'll focus on the most important part: finishing the analysis. Our goal is to master the learning outcome: Interpret the results of a 'Decomposition Tree' to explain factors contributing to a business metric. You'll learn how to transform the visual path into a compelling business story that identifies a root cause and leads to clear, actionable decisions.
From a Visual Path to a Business Narrative
The true power of the Decomposition Tree lies in its ability to structure a root cause analysis in an intuitive way. For the managers in your workshop, the most effective way to frame this is by connecting it to the "Five Whys" method—a technique they may already be familiar with. Each level they drill down into the tree is like asking another "Why?".
Let's explore a practical example that perfectly illustrates this.
Using the Power BI Decomposition Tree for Root-Cause Analysis
The article 'Using the Power BI Decomposition Tree for Root-Cause Analysis' provides an excellent framework for interpretation. It connects the visual directly to the 'Five Whys' method and shows how to translate insights into actions.
Please read the following sections from the article: Start with 'What is Root Cause Analysis (RCA)?' and the 'Five Whys? Method' to ground your understanding. Next, read the 'Use Case' section to understand the business problem: a decline in Customer Satisfaction (CSAT). Now, the most important part: study the '5 Whys for Low CSAT in action' section. Pay close attention to how each 'Why' corresponds to a level in the decomposition tree. Finally, review the 'Resolution' and 'Expected Results' sections to see how the analysis leads to concrete business actions.
As you saw in the article, the process of interpretation is a narrative-building exercise. For the RetailCo example, the story unfolds like this:
- Why is CSAT low? The tree shows the primary driver is
Issue Type: Wrong Item. - Why are wrong items being shipped? Drilling down reveals the problem is concentrated in the
Region: West. - Why the West region? The next level points specifically to
Fulfilment Center: Warehouse B. - Why Warehouse B? The data shows a high volume of complaints came through the
Support Channel: Phone. - Why the phone channel at Warehouse B? The final level reveals the issue spikes during the
Shift: Night.
This clear, step-by-step story, discovered directly from the visual, is exactly what you want to teach managers to construct. The analysis doesn't just stop at identifying "Night Shift"; it provides a complete path that enables targeted action, like retraining specific staff at a specific location.
Interpreting AI-Driven Splits
In the previous lesson, you saw that you can use AI to guide your exploration by selecting "High value" or "Low value". When you teach this, managers need to understand what the AI is telling them.
The key visual cues are the light bulb icon (💡) and the dotted line, which indicate that Power BI has selected the next dimension for you.
This video provides a clear demonstration of how to distinguish between manual and AI-driven paths and what the AI splits mean in practice.
Decomposition Tree - AI Visuals in Power BI
This video from Pragmatic Works clearly illustrates both manual and AI-driven exploration, which is essential for interpretation.
Please watch the following segments: 04:15 - 07:22: Observe the manual drill-down. Notice how the presenter chooses each dimension and how the connecting lines are solid. This represents testing a specific hypothesis. 07:22 - 09:18: Now, watch the section on AI splits. Pay attention to how the presenter uses 'High value' and 'Low value' and how the dotted line appears, signifying an AI recommendation.
Absolute vs. Relative Analysis: A Deeper Look
When using AI splits, there's an important setting that changes how the AI finds "high" or "low" values. You'll find this in the Format your visual > Analysis pane. This is a crucial concept for you, as the facilitator, to understand so you can explain it simply.
- Absolute (Default): The AI looks for the dimension with the single highest or lowest value. For example, it will find the product that generated the most revenue, period.
- Relative: The AI looks for the dimension whose value is highest or lowest relative to its peers in that same dimension. It finds the biggest outlier. For example, it might highlight a new product that has low sales overall (so 'Absolute' mode would ignore it) but is selling 400% better than other new products.
Create and View Decomposition Tree Visuals in Power BI
The official Microsoft documentation explains the difference between Absolute and Relative AI splits. Understanding this will allow you to teach managers how to find more nuanced insights.
Read the section titled 'You can configure the visual to find Relative AI splits'. You don't need to memorize the formula, but focus on understanding the concept from the video game example: why 'Platform' becomes the recommended split in Relative mode even though it doesn't have the highest absolute sales.
Here's how you can explain this to your managers:
- Use Absolute mode when you want to find the biggest contributors (e.g., "Who is our single best salesperson?").
- Use Relative mode when you want to find what stands out or what's unexpected (e.g., "Which marketing campaign is performing best, accounting for its smaller budget?").
Test your understanding!
A manager wants to analyze which of their company's 50 warehouses is the most efficient at processing orders. They have a dataset with "Total Orders" and "Processing Errors" for each warehouse. They put "Processing Errors" in the Analyze field and use an AI split. Which analysis type (Absolute or Relative) would be more insightful, and why?
Show answer
Relative mode would likely be more insightful. A large warehouse will naturally have more processing errors than a small one, so Absolute mode would just point to the biggest warehouses. Relative mode, however, would identify the warehouse with an unusually high number of errors compared to the number of orders it processes, highlighting true inefficiency rather than just scale.
The Dynamic Nature of Interpretation
A critical lesson for managers is that insights are not static. The root cause can change depending on the context. The Decomposition Tree is fully interactive, and its results will update based on filters from other visuals on the report page.
When a user clicks on "Q1" in a bar chart, the entire decomposition tree might recalculate and show a completely different root cause than for "Q2". This is not a flaw; it's a feature that allows for contextual analysis.
Use the Power BI Decomposition Tree to Explore and Analyze Your Data
This video by Avi Singh demonstrates how cross-filtering dynamically changes the analysis, a key concept for interpreting results in a real-world report.
Watch the segment from 06:35 to 08:32. Notice how the AI-driven path changes completely when the presenter selects a year or a month from another (off-screen) slicer. The root cause shifts from 'Color' to 'Category', demonstrating the dynamic nature of the tool.
When you teach this, emphasize that the Decomposition Tree should be used alongside other visuals (like slicers, bar charts, and line charts) to explore problems from different angles.
Conclusion
You are now prepared to teach managers the final and most critical part of using the Decomposition Tree: how to interpret its results to drive business decisions.
Key Takeaways for Your Workshop:
- Frame interpretation as storytelling: Use the "Five Whys" method as a simple, powerful analogy for drilling down through the tree's levels.
- Translate the path into a narrative: Each step in the path is a sentence in the story that explains the root cause of the metric being analyzed.
- Explain what AI splits mean: Teach the visual cues (💡, dotted lines) and the conceptual difference between finding the biggest value (Absolute) and the biggest outlier (Relative).
- Highlight the dynamic nature: Stress that insights are contextual and will change based on filters applied to the report. The goal is to explore, not just find one static answer.
- Always connect insights to action: The analysis is only complete when it leads to a specific, measurable business action.
Preview of the Next Lesson:
We have now completed our deep dive into the Decomposition Tree, a tool for hierarchical, exploratory analysis. In the next lesson, we will explore another of Power BI's AI visuals. We'll learn how to use the 'Analyze' feature to explain increases and decreases in data points, which is perfect for understanding what drives changes in trends over time.