Business Intelligence Network - The Vision for BI and Beyond

A Picture is Worth a Thousand Numbers

Published: August 9, 2007

While there have been advancements in visualization products, it is really up to you to learn how to better apply visualization capabilities.

Visualization seems to be a hot topic of late. I suspect this is fueled by a number of things:

• Vendor acquisition – with Microsoft acquiring Dundas (and last year ProClarity) and Tibco acquiring Spotfire, and XLCubed acquiring MicroCharts
  
  
• New visualization products from business intelligence (BI) heavyweights such as SAS Visual BI as well as niche players such as Visokio Omniscope
  
  
• Increasing integration of advanced visualization capabilities into mainstream BI platforms such as Business Objects Xcelsius (some of whose visualizations have been added to Business Objects Dashboard Manager), MicroStrategy’s Enterprise Dashboards, Hyperion Visual Explorer (OEM’d from visualization vendor Tableau), and Information Builders Visual Discovery (OEM’d from ADVIZOR Solutions)

But most of all, it is a growing acknowledgement that sometimes a picture is worth a 1000 numbers.

Stephen Few, the author of Information Dashboard Design: The Effective Visual Communication of Data, defines data visualization as “technologies and techniques that support the analysis and communication of data using visual media.”¹ Graphs can be a powerful visualization tool. So can bolding and enlarging an important number within a report.

All leading BI tools have basic visualization capabilities: you can take tabular data and turn it into a bar chart, trend line, and so on. They also support some kind of conditional formatting of data: display positive numbers in green, display negative numbers in red and enlarge those with the worst variances. The ease with which these things can be done differs. The built-in “smarts” of graphing engines also differ. I can all too easily build a senseless chart that shows daily sales for three years as bars, when a  line would be more appropriate. Many BI products also use color in ways that baffle me. In graphing sales by state, New Jersey sales in one chart is hot pink, and in another it is lime green. I would assume the software is smart enough to keep dimensions the same color across multiple charts. It’s not. As BI products have been re-architected to the web, some no longer allow me to change the hot pink back to lime green.

Visualization expert Edward Tufte suggests that a tabular display of numbers is better when 20 numbers or less are involved². And yet, I look at the reams and reams of reports with thousands of numbers. In truth, sometimes you do need a precise number – you want the part number, the customer phone number, the charge on your credit card bill. But when you are trying to uncover patterns, anomalies or opportunities, a dense page of numbers is useless.

As an example, take a look at this report.

 

While it may be useful for looking up individual order amounts and dates, it’s impossible to discern any patterns in the data. I could graph some of this data, but it would first require a fair amount of preparation in terms of sorting, grouping and aggregating. In the end, such a chart might reveal the basics: which product categories are selling the most, for example.

To discover the patterns, though, I would need advanced visualization software. Don’t let the word advanced alarm you. At this point in the BI industry, I am using the term advanced to mean:

• Anything nonstandard in BI platforms
  
  
• Visual display that dramatically facilitates discovery and insight
  
  
• Software that helps you apply best practices in data visualization, even for basic visualizations

As an example, I’ve recently begun working with Tableau Software, an advanced visualization package. Other vendors specializing in this segment include ADVIZOR Solutions, Corda Technologies, and Visual Mining, to name a few. Unlike much of my work, my reason for using Tableau was not for a software evaluation. Instead, it was to uncover patterns in a survey for my upcoming book (Successful Business Intelligence: Secrets to Making BI A Killer App). The survey software is completely capable of creating basic graphs. I could easily spew out charts that showed me the percentage of successful projects versus failures. Discovering the commonalities of the successful BI deployments and the commonalities of the failures, though, required much more smarts, or advanced visualization.

Let’s go back to the dense tabular data I showed earlier. I am not familiar with this data and don’t know what’s important. Using Tableau Software, with a few clicks of a mouse, I can more easily see that technology (the green line) is the top selling product category for all regions. Sales seem to have dipped in 2003, particularly in the eastern region. By toggling the quick filters (shown on the right), I can focus on the individual customer segments to see that technology sales to corporations are on a steady decline, whereas consumer and small business segments show strength.

 

Creating this kind of display with standard BI software is theoretically possible with some tools, but one that would take many, many more steps. Tufte refers to this type of graph as “small multiples,” a visualization feature that most BI suites currently lack.

Other forms of advanced visualization include:

• Spark lines – a highly condensed trend line
  
  
• Bullet graphs (a construct by Stephen Few that includes a target indicator within the bar chart)
  
  
• Heat maps that display two variables as different intensifying colors
  
  
• Decomposition tree – something I’ve only seen in Microsoft’s ProClarity that displays each drill-down akin to an ever-expanding organization chart

In addition to creating these advanced visualizations with ease, another characteristic of good advanced visualization software is that it won’t allow you to do silly things (that bar chart that should have been a trend line, for example.) As I’m not a visualization expert, I at first found this frustrating. I wanted to create a simple pie chart. All surveys have pie charts! In Tableau, you can’t, as visualization experts have apparently found that pie charts are difficult to read and ineffective display instruments. I never knew.

I’d like to tell you that improvements to basic visualization capabilities in BI platforms have been steady, and yet, I don’t think that’s the case. As many BI products have migrated to the web in the last couple of years, some have taken a backward step. Let’s hope that Web 2.0 and the maturing of web-based business intelligence will correct this. It would then seem a natural evolution to make advanced visualization (of the small multiple and bullet graphs sort) an extension to any BI platform. And yet, some of these niche visualization vendors have been around for years.

The final step then is us. Business intelligence experts have to learn how to better apply visualization capabilities. As a first step, look at your dense reports: do they scream for some kind of chart? What data can you filter out? Be brave! At least hide some details in the initial display (let users toggle them back if really necessary.) As a second step, learn from some of the experts on which visualizations are most effective. Read Edward Tufte and Stephen Few for starters.

Applied effectively, a picture can help you turn an overwhelming report into simplified insight.

Endnotes:

1. Few, Stephen, Perceptual Edge, BizViz: The Power of Visual Business Intelligence, March 7, 2006
2. Tufte, The Visual Display of Quantitative Information, Graphics Press, 2001, Page 56

Copyright, ASK LLC, July 2007