I found this post in an old folder that I never posted two years ago. But with a few changes, it is just as relevant today.
In my prior posts, I’ve focused on the technical issues of how to gather data into a Power BI Data Model, how to transform that data using calculated dimensions and measures, and how to display the data using the reporting and charting objects built into Power BI Desktop and even a few that are provided by third part tools. However, I need to discuss another important aspect of data analysis. That is determining the best way to present your analysis to different audience types.
The easiest way to classify audience types is by management levels. At the top level are the C-level executives. You might know them better as the CEO, CIO, CFO, COO, and probably several other TLAs (three letter acronyms) that represent the people that set the overall goals or direction of what your organization will be doing in the future. People at this level rely on well-designed charts to guide their decisions. This is not because they cannot interpret tables and matrices. Rather it is because they need to quickly grasp what the data is telling them and how it affects their decisions. This is best accomplished with charts often in the form of dashboards.
When a data analyst is asked to create a dashboard for their organization’s leaders, they may feel that they must somehow fit every piece of information into a single dashboard. But that would be the wrong approach. The correct approach is to first determine what decisions leadership is trying to make and then determine what specific information they need to make those decisions. Each dashboard should then be created to address the information needed to make a single decision. That decision might be to expand the organization in one part of the country while shutting down operations in another less profitable region. The decision might be to add new product/service lines or to decrease the number of similar offerings within a category that are competing against each other rather than against the competition. Or perhaps they need to evaluate difference source of raw materials or to determine where the best location for the new facility should be or maybe what type of advertising works best for different market segments. In any case, each dashboard must focus on a single issue and completely address that issue within the confines of that single dashboard if at all possible.
When it comes to Power BI, the dashboard capabilities exist to create compelling presentations that can be included in PowerPoint slides or displayed directly. The worse thing you can give managers at this level is a boring presentation that included tables and tables of numbers whose relevance to the decision is hard or even impossible to visualize. Each item on a dashboard may have the ability to drill down to a deeper level of detail, but rarely if ever will they need to drill down to the original data source level. It is always limited to summarized data.
Beneath this top level of management is typically a middle layer of management and staff who develop the strategy on how to achieve the goals and implement the decisions of top management. This level also benefits from similar dashboards with data that applies only to their area of concern. Interactivity in these dashboards to drill down and filter data is more important than that needed by top management but still should not rely heavily on table and matrix reports except as a backup to verify the analysis or to verify or explain outlier data. I would suggest that if management at this level regularly insists on getting full table and matrix reports of the data, they may not trust the summarized data prepared for them, or perhaps they just don’t know what the right questions are quite yet and still need to explore different data relationships. Some might refer to this activity as ‘playing with the data’ to determine which strategies make the most sense to achieve the overall organization goals. Keep in mind that once these strategies are finalized, the information defining what dimensions and measures are most important will be used to create the dashboard for top management to guide their decisions. The analysts at this level may be called the strategic data analysts because their focus is in determine what data is needed to support and define the organizations strategy.
Finally, there are what I call the functional data analysts. While they may work in part from a set of requests for data needed by the strategic data analysts, they also spend considerable time gathering as much data as possible and building different data models to test different assumptions. Often these assumptions are based on their general knowledge of the business/service, but sometimes it is just a ‘game’ of playing one set of dimensions against another set of dimensions to see if there is any relationship that best predicts the observed measures. A tool like Power BI makes it relatively easy to test different assumptions to see if a pattern emerges that might predict the measures. Sometimes the best relationships in the data are discovered quite by accident causing the data analyst to exclaim “That’s funny!” Then they will proceed to adjust the dimensions to fine-tune that factors that best model the observed measures. This is what often occurs in the study of customer demographic effects on purchasing of different products/services in different parts of the county. For the functional data analyst, charts are still important in discovering those special relationships, but the tables and matrices that back up those charts take on increasing importance to verify that the observed effect was not an accident.
Ultimately, the source data needed at each of these levels may actually be the same. However, to include all the detailed data in the data models provided to top level managers of the organization may result in analysis that is slow to update. Top management does need as much flexibility in filtering or drilling down through the data using different dimensions. The strategic managers and data analysts have already determined what dimensions they want top management to focus on and can provide dashboards based on summary data for those dimensions rather than detailed data.
Similarly, strategic managers and data analysts do not need all of the data collected by their functional peers. In fact, their function managers and data analysts should have already weeded out dimensions that do not have a direct impact on the observed measures. They have reduced the data model by removing non-essential columns such as a store’s phone number and perhaps even summarizing the data slightly perhaps by including only daily sales rather than each details on each individual sale.
Think about prices on the stock market for example. Looking at the second by second price fluctuations of a stock can be distracting to the decision on whether to buy or sell a particular stock. In fact, looking at the price trend first by hour, then by day, then by month, and finally by year can give you different decisions at each point along the way. If you are a day trader, you probably need to watch the fluctuations of a stock at a finer interval. But if you are long term investor, the decision on whether to buy a stock or sell it depends on its longer-term trend. Of course, whether you are buying or selling, once you have made that decision, you probably would best be served to watch daily fluctuations or even minute by minute changes to time the exact moment to initiate your trade order.
Getting back to a business scenario, we see that there are several trends in the style of data analysis that you may need to consider. Data analysis performed for higher management levels require that:
- the data model must eliminate all irrelevant dimensions.
- the data model must be increasingly summarized.
- dashboards and individual charts should have less interactivity focusing only on relevant dimensions that affect the decision.
- the presentation layer must become increasingly graphical with data presented at a high but focused level to drive home the point they need to make at a glance.
- the need for tables and matrices decreases and can be distracting like the noise of second by second stock price fluctuations.
- the presentation should focus on the one or two points it is trying to make without introducing side issues and should fit on a single screen/slide with no scrolling horizontally or vertically.
- at the highest level, the presentation must stand on its own without support from the data analyst to explain it.
It should be obviously by now that all of these goals cannot be met with a single Data Model much less a single set of reports and dashboards. You must customize the Data Model and the presentation of the data for each audience level. Of course, one way to do this is to base each higher-level Data Model on the data model at a lower level. This method provides continuity. It also provides a single upgrade path as more recent data flows from the lowest and most detailed levels up to the highest and most summarized data levels.
Now one last point. This cannot be easily accomplished by using just the Power BI Desktop version. At the lowest analysis level, Power BI Desktop may be the best way to build the initial data models and ‘play’ with the possible relationships of different dimensions and measures. Depending on the amount of detailed data, this may even require the data to be stored in Analysis Services either locally or within Azure. Keep in mind that calculated columns and measures will update faster in Analysis Services than if stored in local data. But even then, some data analysts prefer to use a data sample in Power BI Desktop to get a ‘feel’ for the data and to explore different visualizations of the data. Only then will they build the final model in Analysis Services with Power BI or Power BI Premium.
While Azure with Power BI Premium might be where your organization ultimately needs to go, don’t be afraid to start small by using Power BI Desktop. Gain acceptance and recognition for what data analysis can provide at each audience level. Build confidence and complexity over time. But get started and do it. It is better to succeed through a series of smaller steps than to fail attempting to immediately implement your grand vision or even worse, to never start at all.