5 steps to smarter data visualization

Many organizations today are taking steps to become data driven by utilizing models, analytics, data visualizations, and dashboards as an integral part of decision making. This is happening in many contexts such as business leaders improving customer experience; technology leaders analyzing agile, devops, and website metrics; and application teams embedding analytics in their applications.

That means many more developers, analysts, engineers, and managers are going to be involved in developing data visualizations and dashboards. With many organizations adopting self-service BI tools such as Tableau and Microsoft Power BI, it’s likely that at some point you’ll want to develop dashboards or you’ll get asked to as part of your priorities.

These tools are easy to learn, but avoid the temptation to design complex data visualizations that don’t get used. You need a design and development strategy and organizations need established data visualization standards.

There’s plenty of advice out there from strategic practices to design strategies to platform-specific standards for Tableau and Power BI. I’d like to share some practical standards that developers should consider.

1. Use discovery dashboards to prep data

In the course of working with data, a developer, business analyst, or data scientist will often develop charts and dashboards that enable a user to review and discover basic insights. These dashboards are needed as part of the development process when you know little about the data, its quality, and what insights may be useful. During this process, it’s important ask some questions:

• Is the data sufficiently useful or does it need to be joined with other data sources to tell a more complete story?
• Is the data relatively clean or does it require some data prep and cleansing before it’s used for decision making?
• Are grouping dimensions, binning measures, computing aggregations, and other analytics needed to simplify the analysis?

These questions are useful upfront; repeat as needed whenever the data requires additional manipulation.

2. Answer the questions of a defined audience

Once you are ready for that first data visualization, consider the following:

1. What question will the visualization address?
2. Who will use it?
3. What will they do with the insights derived from the visualization?

The first question should help you define a title for the visualization. I’ve seen many forget this crucial step, leaving consumers with no understanding of why and how to use the visualization.

The second question aids in defining the level of detail and sophistication. An executive wants performance indicators and trends, whereas a manager often wants to drill down and understand the why behinds the analytics.

The third question helps identify what type of information is needed and whether integration with another system is useful. For example, if your dashboard is analyzing agile metrics from Jira, then allowing the user to click into Jira releases, epics, and stories may be useful. Knowing who the end user is can help determine what types of activities they need to perform with any integrated system.

3. Establish consistent layouts, chart types, and styles

It is tempting to create dashboard layouts that fully reflect all the dimensions, hierarchies, and filters that end users may want to answer their questions and review the analytics.

But do this across too many dashboards and end users will be turned off by the complexities of learning and working across different designs. It’s equivalent to designing a website with different navigational bars on every screen.

That means coming up with consistent layouts, chart types, and styles. Users coming to your new dashboard should feel at ease because of its commonalities with other dashboards they already use. Search interfaces and filters should be in consistent locations. Primary dimensions should use consistent chart types when shown on different dashboards. Dimensions and measures should also use consistent colors, fonts, and formats.

4. Drive storytelling with visual elements  

Once you have a question and have designed the structure of the dashboard based on layout standards, the next step is to make sure the visual is useful and insightful. Don’t expect end users to have their aha moment on their own. The best dashboards use explicit and implicit clues to grab their attention.

There are entire books and many articles on storytelling with data and visual design, but here are some basics:

• Use more visuals and fewer filters for dimensions that drill into data. Visuals provide context through color, size, and trends that are added context versus filters.
• Get the basic charting standards implemented by labeling, positioning, and styling axes, tick marks, color legends, and data points.
• Outliers can be important. Avoid filtering them out and highlight the ones that may be significant.
• Users generally understand basic chart types but may be less familiar with scatter plots, box and whisker plots, radar charts, multilevel pie charts and other more complex chart types. Only use complex visuals when they add value and highlight insights. When using them, reduce other elements on these dashboards so that they don’t overwhelm end users.
• Many BI tools have ways to annotate data or to capture a sequence of dashboard interactions as a story. Learn to utilize these tools. In most corporate settings, being clear and direct is the most practical approach to getting people to see insights.
• Storytelling requires context. Data visualizations can do this with trends, baselines, comparisons, benchmarks, scenarios, forecasts, and other elements that provide the background for analysis or decision making.
• Personalize the experience. For example, if your end user is only interested in European performance metrics, let them set that preference globally so dashboards open with this context.

If this interests you, consider learning more on data visualizations by reading books like Storytelling with Data, getting inspired by top visuals on Tableau Public, reviewing these top data visualizations of 2018, learning from these experts, or following these visualization blogs.

5. Iterate on the data and designs

Once a dashboard is being used, it will need ongoing enhancements.

New dashboards often expose underlying data quality issues, especially when developed from new data sets. The first dashboard versions should address the primary data quality issues that may be barriers to smart decisions.

Data visualizations are like software applications. Usability improvements are another type of enhancement. A dashboard developer should get a minimally viable product in end users’ hands, observe them using it, and collect feedback that can drive improvements.  

Lastly, dashboards also must reflect the current state of decision making and data. Today’s insights become actionable and should become less important over time. New business objectives and stakeholder questions emerge that require analytics and data visualizations. Sometimes this will require creating new dashboards; other times improving existing ones is the more efficient option.

Data visualization practices account for these changes by standardizing version control, testing and release management as part of developing and deploying dashboards.

As more people develop dashboards or participate in business intelligence programs, organizations should establish and improve their development and design standards.