In the last decade data analysis has been moving from the IT areas into marketing , business and operations, through the use of more intuitive tools and data analysis platforms. Within this technology shift, professionals started to build a new language based on visuals that can tell a story in a quick and impactful way. Creating visual representations is not a special skill anymore, it is a must have, and if your only experience with data is an excel chart or google chart, certainly you will need to upgrade your knowledge.
Managers and Executives of successful companies are achieving high levels of Dataviz and encourage their teams to think visually. If employees are not able to transmit their message simply, and make it accessible, probably they are not dominating the visual thinking trend.
In this article we will describe a simple framework for creating meaningful visualizations based on visual thinking principles published by Scott Berinato in his book "Good Charts, the HBR Guide to Making Smarter, More Persuasive Data Visualizations" ISBN 9781633690707. Scott's framework provides a summary of the new emerging visual language patterns and how to tackle in everyday data usages.
Basic Visual Thinking Principles
Before describing the framework, we need to mention the 5 visual thinking principles behind it. They are deducted from the way our brain interprets information.
- We don't go in order
- We see first what stands out
- We see only a few things at once
- We seek meaning and make connections
- We rely on metaphors
A Framework for Visualisations Construction
As many frameworks in business and strategy, the starting point always begins with questions. In this case the framework states that every time you face the challenge of creating a visualisation you need to ask yourself two (2) questions.
- The information I'm managing is conceptual or data-driven?
- My message intention is declarative or exploratory?
Conceptual vs Data-Driven
Conceptual information is derived from the interpretation of an information as concepts, presented around a context. Some clear examples are when you want to describe a "process" or a relationship, no directly related to data points. For example, a Business Model Canvas would be a good example of conceptual visualization, the same as the Technology Adoption Curve.
Data-driven information is straightforward data points. Measures and their dimensions (for those with more math background, vectors) Each measure is under a certain context of measurement, predefined and with its boundaries. A clear example of data-driven information is the average temperature by the state.
Sometimes a merge of Data-Driven and Conceptual information can exist.
Declarative vs Exploratory
This question is not as easy as the previous one, as it has not a yes/no answer. The declarative or exploratory nature of a visualization depends on its purpose.
Business professionals usually work with declarative visualizations. This is because their message is a statement in the form of declaration. Declarative visualisations are designed not to elaborate a conversation, but to generate a discussion. Traditional examples of declarative visualizations would be a sales chart by quarter. The presenter will be showing which where the sale through the time dimension.
When the purpose is to understand or confirm hypothesis over trends and patterns, then the visualisation will be of exploratory nature. This means the presenter will be surfing and drilling the data with different visualisations that try to explain the "why" behind the data.
Potential Types of Visualisations
Based on the previous questions, Scott Berinato created a 2x2 matrix that helps to quickly understand the visualisation that best fit for the purpose of the message and nature of the data. The X and Y axis represent Declarative vs Exploratory and Conceptual vs Data-Driven. Mixing nature and purpose the quadrants become:
- Idea illustration: This quadrant represents the conceptual-declarative corner. Visualisations in this cell describe ideas and concepts, through process and structure diagrams. Examples are Organizational Charts, Layered Pyramid diagrams, Process Diagrams and Funnel Diagrams. Its purpose is simple, translate a complex idea into simpler chunks more understandable. Clear and simple design is important to achieve maximum retention.
- Idea generation: This quadrant is the mix between exploratory and conceptual purpose. This means that this visualisations will try to describe metaphors or trends exploring non precise data. The visualisations in this category are those born in the white boards, where ideas around data are being discussed with the purpose of finding something new. Our brain hypothesizes around patterns and relationships. The lack of precision (not pixel by pixel data) increases the ability to abstract the thinking and find new trends. Examples of this visualisations are draft market share diagrams, approximations of some measure (as sales) and their trend over time, averages of a well know business metric as support tickets, etc. This information will be visually sketched and generate a quality conversation for exploring ideas.
- Visual Discovery: This quadrant represents the data-driven - exploratory merge of nature and purpose. This is where absolute data science converges with statistical inference. This means precise data being drilled up and down in order to confirm a hypothesis, or to derive new hypotheses through patterns inference. When your purpose is confirmation, the visualisation tries to answer if the suspect is actually true, or, if there are other ways to explain the same behavior. In this case, users try to show correlations. The visualisations are traditional lines charts (to find trends, forecasts ,etc) or scatter plots (to find clusters). When the purpose is completely exploratory, this means that the user is not clear in what wants to find, creativity is a key aspect of the process and the modeling activities arise. The visualisations chosen need to force insights more than confirm them. Interactively the visualisation and the user need to traverse several iterations until the data starts providing insights that then become information. Visualisations in this cell are more "sophisticated" as network diagrams with weighted lines, graphs, Sankey diagrams, parallel coordinates, bubble charts, etc. Generally, more than two dimensions are represented in order to find correlations still unseen.
- Management Visualisations: This quadrant is where management focus. Simple visualisations, with a declarative statement. Communicate a simple idea, precisely, backed up with real data. Traditionally used in meetings and presentations, this visualisations are a design challenge not because of their structure, but for the effect they need to generate. The visualisations needs to be auto-explanatory. In this quadrants the visualisations employed are pie charts, line charts, column charts, dashboards, stacked charts, etc.
The Process - Building Better Charts
Creating sophisticated graphical visualisations is today a simple task. Users can gather data from different sources and use it as an input for charting tools. Users can create flat charts, 3D charts, color charts, static charts, interactive charts, etc. Producing a visual appealing chart is easy; but it does not mean that the chart is correct.
Recalling the previous section, a good chart is the combination of design and context. So even though users are most tempted to jump directly to the graphical tools, the first step of the process requires to define the Context.
First Step - Context & Preparation
As in any professional activity, preparation is required. For creating a good chart, users need to abstract their thinking out from the data, and make sure the context is clear to the audience. This does not mean to ignore the data, but to understand its purpose and the purpose of its interpretations. A simple example would be how to see a time sales histogram for an ecommerce site? Traditionally managers would tend to take the data and visualize amount purchased vs time of the purchase. This would be a good visualisation, but more insightful would be to see the sales vs the time of purchase of the buyer. So, users need to have in mind what is they want to show, rather than what they can show. Users can frame their thinking navigating through the four quadrants of potential visualisations and writing down ideas. Another example of context , using the same example of sales by date, would be the fact that sales have seasonality; it is not the same to compare one quarter sales vs another quarter sales if both quarters are not in the same season. The comparison will be correct in terms of data, but its interpretation might be incorrect.
Second Step - Socialize your Solution
As every creative process, sharing ideas to a third party will bring some outsider opinions and fresh comments that will enrich the final result. Talk and listen is an important step; fundamental when the user is extremely knowledgeable of the data and its domain, as generally this scenario generate blinds spots. Sharing the different options of visualisation at this point will allow the user to select those charts that are easy to interpret, within the context and the message they need to transmit. Data is not key at this point, the questions being asked are more important:
- What I'm trying to say?
- What I'm trying to prove or learn?
- Why I'm showing this data?
Third Step - Sketch your ideas
An Image worths a thousand words. Now is time to pick up the different visualisations available and sketch them. This step will allow the user to visually understand how to produce results from the visual object being referenced. Different sources of data trying to answer different questions within different contexts, require different visual approaches.
After socializing the ideas, the user can look up at the keywords brought up in conversations and sketch them over the charts. Starting with abstract charts and evolving them through iterations. At this point just let creativity fly.
If the user is not really familiarized with the vast options of visualisations, it can base its initial decision in the Abela's Chart Type Hierarchy, a simple guide with chart suggestions depending on the data structure and its purpose. The cheat sheet will not fit all purposes (more sophisticated data or context will require more imagination or knowledge) but generality applies for this kind of tools; a high percentage of cases will fall under the useful tools category and will be able to take the most of the cheat sheet. In the minority of cases, the guide might narrow the options, so use it wisely and do not abuse.
Sketching may take within 15 minutes to an hour. The experienced users will sense when it is enough. Is necessary to set a boundary in order to avoid information paralysis; after an hour, stop sketching and move to the next step with the actual outcome. Some insights that you've done sketching:
- The sketches are a reasonable representation of "What you are trying to say"
- Sketching Iterations are becoming refinements.
- You are already trying to put data into the sketches.
- You are thinking in the look and feel of the chart.
Fourth Step - Prototype
The prototyping step consists of iterations of refinements over the selected visualisations. Prototypes should be created based on real data. Often, it's useful to prototype on the basis of a small subset of the data to create accurate pictures without feeling the burden of having to prototype everything. Prototypes should also include design decisions, such as theme and colors that will define the visualisation look and feel.
The audience context should be analyzed during prototyping. Is the audience expecting "interactivity" with the visualisation ? Is the audience expecting static content ? Are the visualisations going to be published printed or online ? This questions will help the user to prototype different user experiences in order to test with real audience representatives.
During prototyping, if the data is vast, and the tools complex, it is wise to combine the voice of the subject matter expert of the data, the tools expert and visual designers. This profiles combined will define the ultimate visualisation based on the user's sketches.
Prototyping can occur in whichever tool the user plans to use for the final delivery. Ranging from paper to digital tools as excel and complex tools as business intelligence engines. Digital tools for charting are exploding, options are wide. In the last section of this article, we propose some options to create stunning designs.
Loop - Always Refine
The process described never ends in the fourth step. Once the charts are created, users can loop over and over refining several aspects of their visualisations providing the audience a better and cleaner experience.
In order to take advantage of the visualization framework we include in this article three categories of valuable and popular resources the reader can use to create astonishing visualisations that will impress their audience.
PowerPoint and Excel
Almost every business PC in the modern workplace is equipped with Microsoft Office productivity suite. PowerPoint and Excel are the entry level tools for professional visualisations. Both tool are extremely powerful, but also, can be misleading if used improperly. If the user is not experienced with the tools, it will end up creating built int charts, with default styles and underachieve in the visual impact. These tools are great for sketching and prototyping. Once the user knows the kind of visualisation that wants to present, is time to move to another tool, or use professional business templates that will help. For Excel, you can find useful Excel Templates at Vertex42, a popular site with thousand of resources for charting and presenting information. For PowerPoint you can find PowerPoint Templates at SlideModel, a professional powerpoint library with thousands of professional decks designed for presenting awesome visualisations, charts, dashboards and infographics. Both resources are aimed for business, and created with data driven designs. Users can edit the templates and complete them with their data, fill the placeholders and create awesome visualisations in few clicks.
Self Service Business Intelligence
As we stated in the introduction, the modern workplace is evolving into data driven organisations, where decisions are made over trusted data. Software companies have envision this scenario a decade ago and invented the business intelligence term, and their software suites. Business Intelligence stands for multi dimensional models, summarized data, Key Performance Indicators (KPI's) and Reports. But the concept of BI is extremely hardcore and delegated to IT areas. For this reason, software companies are pushing analysis areas to take control over data and create their own vision and reality over their findings. Under this idea is that Self Service BI tools have been created. Simple and Powerful tools that allow regular PC users to interact with data in a multidimensional fashion and with powerful visualisation tools, just accessing the data sources, without centralized control.
The first tool that appeared in the market for self service BI is Excel and its popular Add-In PowerPivot. PowerPivot extends Excel into a powerful BI tool, adding new visualisations as gauges,Maps, Tree Maps, Histograms, etc. Following the trend a new startup grew fast, called Tableau. Tableau quickly became a popular tool due to its charting featured. It included much simpler and powerful charts than Excel, and a simpler way of creating and sharing them. As a competitor of Tableau, QlikView was born. It used the same in memory tables as Tableau, and added some additional visualisations. Finally, due to the surge of these tools popularity, Microsoft evolved PowerPivot into a new Product called PowerBI. PowerBI not only provides reporting and visualisations, it also provides a cloud service for publishing these dashboard in the web.
One of the features that all tools share,and make this tools so powerful is the ability to be interactive. Users can interact with each visualisation, and the rest of the visualisation in the same context responds to this interaction. This feature known as "slicing" allows users to create set of visualisations and flows of slicing paths, that show insights across different variables.
Statistical Visual Tools
The previous tools described are excellent for business but when the domain is strictly scientific, more powerful visualisation tools are required. We are talking about statistical inference and visualisations that need to consider millions of datapoints. For this kind of multi-vectorial data specific tools exists. The most popular within academia and data scientists is known as R. R is a software and environment for statistical processing and graphics. Users can create amazing well designed, publication quality visualisations, with high accuracy. The project is open source, the documentation online is vast and the industry is spreading its use from academia to business.
In this article, we reviewed a proven four-step framework for creating meaningful and useful visualisations. With the concept of visual thinking behind each step of the framework, users can apply the best practices of visual design to their charts and focus in the persuasion and not only in the information.