Rashomon Effect — Lessons for Building Effective BI Dashboards
This image of the elephant is commonly used to illustrate the impact of Rashomon effect on business decision-making. The idea here is that each person offers a different perspective because they are only seeing a part of the full picture. And when a set of facts is too large for any one person’s grasp, then each person will have an incomplete picture and likely draw varying conclusions. In the context of business intelligence, this means providing the end user a result set that’s small enough to analyze and interpret.
Bottom line: Do not create charts that have more than 2–3 data dimensions.
Choose the Simplest Possible Visualization
Stick to the major well understood chart types: line, column, bar et al. If you need to show more than two dimensions, consider using a plain data table or some commonly used chart types such as a bubble chart. As BI platforms continue to innovate on new ways to visualize data, its all the more important to stick to the tried and tested especially when designing for a diverse groups of non-technical stakeholders.
Bottom line: Stick to chart types available in Microsoft Excel. If a chart type is too innovative for Excel, then its likely to be too complex for the end user.
Position & Sequence Charts with Care
Once you’ve designing the charts, start arranging the charts to match the business objectives. Typically, people build dashboards in layers, with the most important high-level KPIs on the top and the layers below providing greater details and drill downs into the high level KPIs. It is also a common practice to start with the most frequently used charts on the left and way to the right in order of decreasing frequency of views.
Whatever your approach, its important to remember that each chart in a dashboard is a part of the whole and the sequence in which information is presented affects the user’s interpretation of the meaning behind the charts.
According to Pew Research, the order in which questions are asked in a survey tends to have a significant impact on the responses (“order effect”). Similarly, the interpretation of later charts in a dashboard is impacted by the user perception to the preceding charts in that dashboards. While its important to let business questions influence dashboard design, it is critical to consider the consequences of unintended bias being introduced through the process.
Bottom line: Do not have more than 4–6 charts in a dashboard; the later charts are least likely to be useful and can cause unintended confusion.