Visualizing Business Intelligence: from Spreadsheets to Insights

What is Data Visualization?

Data visualization is a practice that has great potential and should be taken seriously.  Contrary to popular belief it is not an extension of visual design, but rather an extension of the field of statistics.  Although it INVOLVES both fields, communication must be prioritized over beauty; otherwise you have data art and not data visualization.  This practice has an enormous power to unlock insights for people working in virtually any field that involves the interpretation of data.

The utility of doing it the right way is realized most when dealing with professionals that need to see their data clearly in order to make good business decisions.  In these scenarios it’s not acceptable to deliver dashboards and reports full of visualizations that are cute yet useless.  I’ve witnessed many scenarios where people get dazzled in the demos and disappointed with the delivery.  It’s best to avoid that!  The goal of data visualization is not to take data and just express yourself with it.  The goal is also not to create the most eye-catching image you can (unless you’re designing only for demos).  The goal isn’t to take data and to see what can be visualized with it.  Instead, the goal is to first decide what the target user needs to know, then to design the visualization so that it best suites that purpose.

Thankfully there have been a number of profound thinkers and authors (some of whom I’ve directly had as my teachers), who have enlightened the world on how to effectively visualize data.  My approach is inspired by the work of Edward Tufte, Stephen Few, William Cleveland, Naomi Robbins and Alberto Cairo.

The image shown here is an example of a visualization I quickly mocked-up to roughly illustrate my process in creating data visualizations.  A graph is best built by understanding the dimensions of what needs to be displayed, its format, and features that will promote insights.  Finally, a polished visual design is needed to make sure the graph is fully effective.  Data visualization is the practice of using what we know about the human sensory, perceptual and cognitive systems, and leveraging this knowledge to display data in a way that communicates, informs and inspires insights.

Determining the Best Data Visualization Approach

When creating visualizations, it’s important to understand the role of the person who will be consuming the data.  We need to know that person’s goals, metrics by which their performance is measured, their motivations, what they’re accustomed to, and most importantly: what decisions they need to make with the incoming information.  If I can get that information, I can create a nearly perfect visualization for a person in virtually any job role (the next slide discusses how I validate this).

The images presented here are low-medium fidelity mockups that resulted later in a more polished and high fidelity deliverable (which won’t be shown here for proprietary reasons).  The first is a dashboard for a job role responsible for analyzing the profitability of oil shipping fleets.  It contains data filters and line graphs with actual and projected performance.  There are considerations behind every aspect of these graphs, such as the choice to use multiple graphs, the amount of information displayed on a scale, the careful use of colors (ensuring that when colors are used they are noticeable and meaningful), and so on…  The second image is a tool for a job role responsible for analyzing the quality of fuel obtained from different sites around the world.  My research showed that these analysts were mostly interested in four fuel-attributes and wanted to be able to directly compare the different fuels.  When this visualization was tested, it was clear that this analysis became much easier for these individuals.

My top priority is for the visualization to actually work, and to ensure this requires testing.  Testing a visualization doesn’t mean you just throw it in front of the target user and ask: “what do ya think?”  Instead, you want to watch the target user actually use the visualization in a meaningful way.  This concept actually isn’t novel at all; it’s really the same thing as any valid usability test, yet people rarely think to do this prior to delivering their data visualizations.

Verifying Visualizations

To make this test (mentioned on the previous slide), you compile a set of questions reflective of things the target user would actually have to do with the data.  So for example, if you’re testing a sales analytics dashboard, you might ask: “Which sales channels are bringing in the most earnings?”, “How do the different regions compare to each other?”, “Is the marketing effective?”, “What do you think is going to happen next quarter?”

The image shown here (across, in the upper right) is from a visualization I did a long time ago, as an entry into a visualization contest. (I’m telling this story to describe my hopeful path towards expertise).  There were 90+ competitors worldwide, and the task was to visualize a dataset of a school teacher’s math class.  Everyone worked on the same dataset.  The contest was judged by an expert, Stephen Few.  I did not win the contest, but it was fun and I learned a lot!  What I noticed is that my visualization was very similar to the winner’s, in terms of overall structure, choice of graphs, etc.  The winner and his team had a much more polished visual design.

The reason I’m discussing this example is because I actually tested my visualization prior to submitting it.  I tested it once on an actual school teacher (which was the target job role).  Then I modified the visualization and tested it again.  The test consisted of a series of questions of interest to school teachers who are reviewing their classroom data.

This dashboard is essentially a list of students, and a set of columns showing the different aspects of their performance.  It consists of bars, lines, a heat map, and a frequency polygon.  The interesting thing is that although this visualization appears somewhat complex at a glance, when it was put in front of the target users (school teachers) they actually thought this visualization looked simple (despite lacking expertise in data analysis, statistics, etc.).  The reason this visualization was effective is because it presented the right information in the right way.

Specifying Visualizations for Developers

When the design is finished, it’s important to make sure that the actual product is built correctly.  In some ways, this is plain old UX design; but in other ways, data visualization is an area that has some unique challenges.  When spec’ing for data visualization, I always make sure the developers understand that some aspects of the design are less robust to compromise than others.  In other words, if you cut corners on the wrong stuff it can ruin the visualizations and render them useless in the real implementation, despite whatever beautiful wireframes may have preceded it.

The image shown here is a wireframe for a dashboard I designed for a data governance job role (“data custodians”).  It was part of a larger project where a particular company was merging databases from their different departments, all into one new database.  This visualization succeeded at displaying almost 200 data points on one screen.  The target users found it easy to read and relevant for their jobs.  The specs that I wrote for the developers contained this image along with detailed annotations about how each portion of the dashboard needed to function.

In closing, data visualization is a field that has great potential to improve people’s work lives, when done correctly.  You want to hire a professional who is educated in statistics, understands how the human perceptual and cognitive systems work, has a good process for determining what kinds visualizations to create, tests their designs, and specs it out well for the development teams.  Following this methodology is a recipe for creating some really impactful work.

Back to top