The fundamental question that all scientists — from elementary science students to NASA engineers and PhDs — aim to answer has evolved little since the early philosophers began to question the world around them. Evident in the ongoing babble of toddlers exploring their environment with new eyes, it is human nature to want to know “why.”
This curiosity doesn’t leave us as we grow up; rather it morphs and evolves as the scope of our problems changes. In business, we don’t ask our teams why the sky is blue, but we do ask why a certain combination of strategies is the best approach to achieve our desired goals. We start with “why,” plot the best course of action, track and analyze KPIs and adjust based on the insights we find, before we do it all over again. In our ever-faster-moving business environment, executive leaders strive for a clear understanding of their business data, and to digest it quickly and execute strategies without slowing innovation. But this process cannot happen without support from data-savvy teams.
As businesses mature in their analytics journeys, their teams should evolve to present data in succinct ways that make sense for the context and message of the information being conveyed. In order to help business practitioners understand when it is appropriate to use which type of data visualization, we will break down each data visualization type. We will also explain when is the best time to implement it as you build a dashboard and strengthen your visual vocabulary — all in the context of distinguishing between decision boards and dashboards.
This practice is not limited to data science-heavy industries and verticals. CIOs, CFOs, CMOs and even Chief Data Officers can benefit from improving the way their teams present and how they interpret data.
To understand how to work toward implementing decision boards we have to understand where we started: dashboards. By now, we are all too familiar with analytics dashboards, which include the default integrated reporting platforms of the digital tools we know and love, such as Google Analytics and Hubspot. They are effective at providing a high-level snapshot of performance broken down by category (day of the week, location, age, gender), and they are visually appealing but require a presenter who puts the data in context to answer the fundamental question: Why does this matter?
Decision boards, on the other hand, are fluid. They aggregate the data from cross-organizational channels to paint a clear, easy-to-follow picture that goes beyond descriptive metrics. These are often custom builds designed for an organization’s specific needs. Varying by the level of analytics maturity and design resources, decision boards can also illustrate diagnostic metrics, or why something happened; predictive metrics, or what is likely to happen; and prescriptive metrics, or what needs to happen next. Making the jump from dashboards to decision boards requires basic knowledge of design thinking, which when integrated into an organization’s culture can advance its analytics and reporting capabilities.
The most effective decision boards are created when we implement design thinking. Loved by corporate powerhouses like Google and Apple, and legacy academic institutions like Harvard, design thinking’s methodical process means we get to the heart of the problem quickly, every time. It is efficient and built around the people who will use it — two staples of the insights we are trying to build. As part of design thinking, teams can assess which of the four major metric types (or combinations thereof) are needed to build a decision board.
- Descriptive Metrics: Though not inherently valuable for decision-making, descriptive metrics give a snapshot of what has happened or is currently happening. They are a real-time glance at how multiple variables work together. Graphs and charts that illustrate descriptive metrics include:
- Distribution (box plots, histograms, dot plots)
- Part-to-whole (pie charts, waterfalls, stacked column charts)
- Correlation (scatter plots, XY heatmaps, bubble charts)
- Diagnostic Metrics: Diagnostic charts allow decision-makers to ladder down from the descriptive metrics to the “why.” In decision boards, diagnostic charts are linked to their correlating descriptive metrics, so that users can logically draw conclusions when they click on the data. Displaying diagnostic information is more about the flow of data than the structure of the chart. When choosing what graph to use, it is important to evaluate what specific questions you are trying to answer. The following structures are most often used for diagnostic charts:
- Flow (chord diagrams, networks, Sankey charts)
- Distribution (barcode plots, cumulative curves, population pyramids)
- Predictive Metrics: Perhaps the simplest to understand, predictive charts forecast what will happen based on the existing dataset. These metrics are critical in making the transition from dashboards to decision boards and, when done correctly, should chart a clear path to the next steps.
- Correlation (line+column, scatterplot, bubble chart)
- Change Over Time (line chart, connected scatterplot, area)
- Deviation (diverging bar, surplus/deficit)
- Prescriptive Metrics: The divergence into prescriptive metrics tips the scale from dashboards to true decision boards. These displays of data indicate the next steps for business leaders. Requiring the most advanced data science knowledge, these charts use AI and ML to optimize performance.
As you build decision boards, focus on flow. Think about how your information will be digested and aim to create the most logical structure for your boards. This is where the basic principles of UX/UI design will benefit your teams the most.
The learning curve for building charts can be difficult, but not so difficult that a general business user can’t get the hang of it with time. To help with the construction of your decision board, LatentView has created a Visual Vocabulary, which is an open-source guide to building custom charts in Tableau. Periodically, LatentView will release step-by-step tutorials that walk users through employing Tableau filters. The first installment covers data source and extract filters.
As your company progresses on its data analytics journey, there are a few key pillars to remember. First, make your decision boards easily accessible to the right stakeholders. Done well, these boards serve as an ongoing resource that is meant to be accessed regularly rather than presented at quarterly meetings. This is the primary reason decision boards are a more effective tool than previous iterations of data visualization.
Second, continue to ask for feedback and refine the structure of your decision boards. The composition of your boards will evolve as your business needs do.
Finally, be relentless in your pursuit of the “why.” It will make your predictive charts stronger, more intuitive and more sustainable in the long run. And by the way … the sky is blue because the gases of our atmosphere refract white light from the sun, scattering blue light waves (the shortest and quickest of the color spectrum) across the daytime sky.
Chart types index
Bubble chart: Gives us a glimpse of the current state of the business. This chart provides an overview of sales (on the y-axis) against profit (on the x-axis) for varying subcategories. The size of the bubble is proportional to the size of the sale and the color represents the respective category that each subcategory belongs to. A quick glance shows that the subcategory “Tables” is on the lower side of profit despite a reasonable number of sales.
Waterfall chart: Another way to exhibit positive and negative factors that affect the total profit, broken down by subcategories. Using the key as a guide, the sample below shows that the “Bookcases” and “Tables” subcategories are largely responsible for profit loss.
Sankey chart: Visualizes the flow of data. In the waterfall chart example above, we observed that both sales and profit for categories that fall under “technology” were higher as compared to other office supplies. To understand the major contributors to this category, the next chart clearly shows that phones and machines are responsible for the majority of sales. (Note: The width of the arrows represents the magnitude of the metric under discussion.)
In the below snapshot, the quarterly sales show an exponentially increasing trend over several years. It is also good to know what the future trend could look like. Hence, the forecast chart plays an invaluable role in certain cases. The prediction of sales will help businesses estimate factors like allocation of resources or expanding markets.
Cluster chart: Helps us understand different types of clusters formed based on Tableau’s backend k-means algorithm. With sales vs. profit illustrated below, cluster 1 depicts low profit and low sales typically with the most number of data points; cluster 2 depicts moderate sales and profit; and cluster 3 depicts maximum profit and sales. Further drill-down analysis of the cluster 1 data would lead to clarity for further action needed, like how to improve marketing strategy or financial management.
Author: Boobesh Ramadurai is the director of data and analytics at LatentView Analytics.