The power of AI in Power BI: A data scientist’s perspective
As a data scientist, I often work with and analyze large, complex datasets and then communicate the results to various stakeholders. This can be both challenging and time consuming, especially when I need to create reports and presentations that are clear, concise, and engaging. That’s why I have been excited to discover Power BI AI features, a set of powerful tools for helping to create interactive and insightful data visualizations and narratives using Artificial Intelligence (AI) and Machine Learning (ML). In this article, I share some of my work experience and lessons I have learned using Power BI AI features, and how they have helped me improve my data science skills and productivity. I focus on four features that I have found particularly useful: Power BI CoPilot, CoPilot Narrative visual, Q&A, and Key Influencers.
Power BI CoPilot: A smart assistant for data storytelling
One of the features that impresses me the most about Power BI AI features is CoPilot, an AI-powered assistant that helps me create summaries, insights, and suggestions based on my Power BI data and reports. CoPilot is a preview feature that you can enable in the Power BI desktop or service, and it works by analyzing your data model and visuals and then generating natural language narratives that highlight the key points and findings. You can use CoPilot to create a report page, a summary, or a custom narrative based on your own prompt. You can also use CoPilot to suggest new visuals or questions that you can ask to explore your data further.
For example, I can use CoPilot to create a report page based on a semantic model that contains fabricated sample data about Azure Sentinel, a cloud-native security information and event management (SIEM) solution. Suppose I want to understand how different customers, services, and incidents perform over time, and what factors influence the Azure Sentinel ACR, a metric that measures the quality of security alerts. To use it, I simply click on the CoPilot button and select the suggested content for this report option. CoPilot then creates a report page with eight visuals and a filter, as shown below:
CoPilot quickly and accurately creates this report page, providing a good overview of the data and insights. I can see that Azure Sentinel ACR has fluctuated over time, influenced by several factors including the category, severity, and number of incidents. CoPilot also adds some tooltips and labels to the visuals to make them more informative and user friendly.
Of course, the report page is not produced perfectly, leading to some of my own adjustments and refinements to make it more suitable for my audience and purpose. For instance, I have changed some of the titles and labels to make them clearer and more descriptive, I have added some slicers and filters to allow more interactivity and customization, and I have tweaked some of the colors and formats to make the visuals more appealing and consistent. But CoPilot saves me a lot of time and effort in the initial stage of creating the report, while giving me some ideas and suggestions that I might not have thought of myself. It also helps me prepare for some of the questions and feedback my stakeholders and leaders might provide, while it points out some key takeaways and trends for me to highlight and explain.
Using the CoPilot Narrative visual
In this section, I share how I have used the CoPilot Narrative visual, which is a new feature that allows you to generate natural language summaries of your data and insights. This is different from CoPilot report page creation as I have outlined above, which provides suggested visuals and layouts for a report. The CoPilot Narrative visual enables you to write your own prompts or use predefined ones to get a text-based analysis of your data, which can be useful for highlighting key points, preparing for presentations, or creating reports for different audiences.
For my report on Azure Sentinel data, for example, suppose I want to create a narrative that summarizes the data and highlights the main findings and recommendations, while being suitable for a technical audience. To use the CoPilot Narrative visual, I add it to my report page from the Insert tab and chose the CoPilot preview option, which enables the AI features of the visual. I have the option to choose specific pages or visuals, but because I want to get a comprehensive overview of the data, I select all report pages for this narrative summary prompt.
The prompt is a set of instructions that tells the CoPilot what kind of narrative I want to generate. For this one, I write “What are the important findings to share with the audience about this data and why? Write the summary in user-friendly terms so that anyone can understand the outcome. Include ideas for what actions we should take based on the information of the data and why. Suggest some key visuals for this data.” I could also use one of the CoPilot-suggested prompts, but here I want to have more control over the output.
The CoPilot analyzes the data and the prompt, and then displays the summarized narrative below:
The CoPilot Narrative visual is a helpful tool that can help you create natural language summaries of your data and insights. It’s not perfect, but it is a great first draft for summarizing key insights and recommendations. I think it can be an effective way to complement your visuals, to communicate your findings more clearly, or to prepare for different scenarios and audiences. You can also customize your prompts and data sources to get the best narrative for your needs.
Q&A: A natural way to explore and query your data
Another feature that I find very helpful and convenient in Power BI AI features is Q&A, a natural language interface that allows me to ask questions about my data and get instant answers in the form of visuals or tables. Q&A is not a new feature, but it has been improved and enhanced with the help of CoPilot, which can help me train and customize the Q&A model to better understand my data and my queries. Q&A is available as a visual that I can add to my report, or as a button that I can click on the top of the report page.
For example, I can use Q&A to explore and query the same Azure Sentinel data that I used for the CoPilot report page. Suppose I want to know more about the distribution and the performance of the different customers, services, and incidents. I can simply type or speak my questions in the Q&A box, and Q&A generates the appropriate visuals or tables to answer them. Some of the questions that I can ask here include:
- What is the average Azure Sentinel ACR by customer?
- Which service has the highest number of incidents?
- How does the incident resolution time vary by severity?
- What is the correlation between the number of AAD logs and the Azure Sentinel ACR?
Q&A understands and answers most of my questions, while also giving me some suggestions and tips on how to improve or refine my queries. For instance, it shows some synonyms or alternative terms that I could use for my fields or values, it offers some examples of questions that I could ask based on my data, and it teaches me some tricks and operators that I can use to make my queries more specific or complex. Q&A also allows me to provide feedback on the answers that it provides, and to teach it new concepts or relationships that it might not have learned from the data model.
I am impressed by how easy and intuitive it is to use Q&A, and how it enables me to explore and analyze my data in a natural and conversational way. I can ask any question that comes to mind and get immediate visual answers on which I can further drill down or filter. I can also save the answers that I like, and add them to my report as visuals, or pin them to a dashboard for quick access. Q&A helps me discover new insights and patterns that I might have missed or overlooked, and it has also helped me validate or verify some of the findings that I obtained from the CoPilot report page.
Key Influencers: A powerful way to identify and explain the drivers of your data
The third feature that I want to share is Key Influencers, a Machine Learning visual that helps identify and explain the factors that influence or drive a certain outcome or metric in my data. Key Influencers is a feature that I can add to my report as a visual, and it works by analyzing the correlation and the impact of the different fields or values on the target field or value that I want to understand. Key Influencers can also perform clustering and segmentation on my data to reveal the top groups or categories that have the most influence on the outcome.
For example, I can use Key Influencers to understand and explain what influences the Azure Sentinel ACR, the same metric that I used for the CoPilot and Q&A features. In this instance I want to know the factors or variables that have the most positive or negative impact on ACR and how they vary across different segments or clusters of my data. So, I simply add the Key Influencers visual to my report, select the Azure Sentinel ACR as the Analyze field, and then I select several other fields that might be relevant or interesting as the Explain By fields. I also select the Categorical option, as the ACR is a categorical variable with three possible values: Low, Medium, and High.
Key Influencers then generates a visual that shows the key influencers and the top segments for the Azure Sentinel ACR, including the following:
- The Key Influencers tab shows the fields and values having the most influence on the next predicted ACR, ranked by their weight or significance. For instance, it shows that the number of incidents created has a negative influence on ACR, meaning that when the number of incidents increases, ACR is more likely to decrease. It also shows that the service Azure Monitor has a positive influence on ACR, meaning that when the service is Azure Monitor, ACR is more likely to increase. Here’s an example showing one of those key influencers:
- The Top Segments tab shows the clusters or groups of data with the most influence on ACR, based on the combination of fields and values. For instance, it shows Segment 6 as having the lowest next predicted ACR, $499 units lower than the overall average while containing nine percent of the data. It also shows the characteristics of this segment, such as the average number of incidents, the most common service, and the most common category.
In this way, the Key Influencers visual is powerful and insightful, helping identify and explain the drivers of Azure Sentinel ACR. I can see the relationship and the impact of the different fields and values on ACR, and how they vary across different segments or clusters. I can also drill down or filter the visual to focus on a specific value or segment to see how the key influencers and the top segments change accordingly. As a result, Key Influencers has helped me understand and interpret the data in a deeper and more meaningful way, while also helping me generate and test some of my hypotheses and assumptions about the data.
Conclusion
This article presents some of the Power BI AI features that I have used and learned from in my work as a data scientist. I have reviewed four Power BI features that I have found very helpful: Power BI CoPilot, CoPilot Narrative visual, Q&A, and Key Influencers. These features have allowed me to use Artificial Intelligence and Machine Learning to create engaging and informative data visualizations and narratives, to ask and answer questions about my data in a natural and straightforward way, and to find and explain the factors that affect or determine a certain outcome or metric.
These features have also enhanced my data science efficiency, as they reduce time and effort, provide tips and recommendations, and help me uncover new insights and patterns. I hope you have found this article interesting and useful, and I invite you to test these features for yourself and see how they can assist you with your own data science projects.
Amy Gillespie is on LinkedIn.
