Whi

te P

aper

Powerful Visual Techniques for Exploring and

Analyzing Quantitative Business Data

Visual and Interactive AnalyticsFulfilling the Promise of Business Intelligence

Stephen FewPrincipal, Perceptual Edge

About the Author

Stephen Few has 24 years of experience as an IT innovator, consultant, and educator. Today, as Principal of the consultancy Perceptual Edge, Stephen focuses on the use of data visualization for analyzing and communicating quantitative business information.

We supposedly live in an “information economy,” yet businesses, the brokers ofthis economy, rarely serve as good models for the effective use of information. Fartoo often, business decisions are made without sound thinking to understand theavailable information and without clear communication of its message to decisionmakers. This problem undermines the ability of businesses to operate effectively,yet it is rarely diagnosed and almost never treated. Stephen is working to raiseconsciousness and to provide a treatment plan that addresses the needs of busi-ness in the language of business. His book, Show Me the Numbers: DesigningTables and Graphs to Enlighten, is a powerful fitness program designed to targetthe data presentation aspects of this problem. His new book due out in January2006, Information Dashboard Design: The Effective Visual Display of Data, appliesvisual design practices specifically to the challenging task of displaying a greatdeal of disparate information on a single screen in a way that communicatesclearly and efficiently.

Today, from his office in Berkeley, California, Stephen provides consulting andtraining services, writes the data visualization newsletter and Blog for the BusinessIntelligence Network (www.B-EYE-NETWORK.com) as well as frequent articles forDM Review and Intelligent Enterprise, speaks frequently at conferences, andteaches in the MBA program at the University of California in Berkeley. More information about his current work can be found at www.perceptualedge.com.

ii

1

The BI industry often loses sight of this clearvision. In many ways, BI is still a fledgling industry,awkwardly struggling with good intentions tomature beyond adolescence, past the flexing andpreening of raging hormones, to the responsiblesolution provider that it has always strived tobecome. The time is right for BI’s rite of passageinto adulthood. Some software companies, likeSpotfire, are showing the way. Some companies(I’ll resist the temptation to name names) are stilltrying to get by on their good looks, flirting withthe sad possibility of never growing up.

The “I” of BI—intelligence—can only be achievedby fully engaging the half of human-computerinteraction that possesses intelligence: the humanhalf. BI is only as effective as its ability to supporthuman intelligence. This requires software thatseamlessly interacts with our brains to supportand extend our cognitive abilities. Unfortunately,BI software too often gets in the way, interruptingand undermining the thinking process rather than complementing and extending it. When BIsoftware does its job, however, you find yourselfsubmerged in thoughts about the data, not aboutthe software and the hoops you must jumpthrough to reach insight.

(Source: A glossary on the Web site www.gartner.com)

Information technology hasn’t delivered what itpromised us. Yes, we live in the information age,and yes, much has changed—but to what end?Do you know more today than before? Are yousmarter? Do you make better decisions? Weoften still make the same bad decisions, but nowwe make them much faster than before, thanksto technology’s questionable gift of “more andfaster”. This is hardly the better world that weimagined and hoped for.

For many years I’ve worked in the business intelligence (BI) industry. It is BI’s mission to helpbusinesses harness the power of information towork smarter. Intelligence—”the faculty of under-standing” (according to the Oxford EnglishDictionary)—is the solid ground on which busi-nesses must build to succeed. Information is thestuff with which intelligence works to produce theunderstanding needed to effect change, but moredata delivered faster can actually lead to lessunderstanding and even bad decisions if we lackthe skills and tools needed to tame and makesense of it. The BI industry has helped us buildhuge warehouses of data that we can now accessat lightening speeds, but most of us look on withmouths agape, feeling more overwhelmed thanenlightened.

The Gartner Group coined the term businessintelligence in the mid-1990s and defined it as follows:

An interactive process for exploring and analyzingstructured and domain-specific information to discern trends or patterns, thereby derivinginsights and drawing conclusions. The businessintelligence process includes communicating findings and effecting change.

An interactive process for

exploring and analyzing

structured and domain-specific

information to discern trends or

patterns, thereby deriving

insights and drawing conclu-

sions. The business intelligence

process includes communicating

findings and effecting change.

Visual and Interactive Analytics

2

As a business person trying to get the job done,your only concern when dealing with data is:“what does it mean?” so you can move on thewhole point of the matter, which is to decide:“what should I do about it?” That’s it. Did youknow that there are experts who spend all theirtime figuring out the most effective ways to makesense of information to support good decisions?It’s true. While you’re sitting in your office or cubicle doing your job, they are sitting in theiroffices or cubicles in research laboratories, most at universities and a few at commercialenterprises, doing their best to help you. Theirfindings get published, but rarely anywhere thatyou would ever see.

This is where people like me come in handy.I try to keep one foot planted firmly in the business world and the other in the researchworld so that I can pass useful information backand forth between the two. Researchers musthear what you really need down there in thetrenches of everyday business, and you needtheir insights without all that high-fallutin’ academic speak that keeps too many goodideas trapped in the ivory tower.

I was recently involved in an email discussionwith my friend and colleague, Ben Shneiderman,of the University of Maryland. Ben is not just anyold academic involved in information visualizationresearch; he is a prime mover in the field. No onehas done more to further this work and promoteits worth. Ben and I have been exploring ways tostrengthen the bridge between the informationvisualization research community and commercialbusiness software. We realize that if informationvisualization sounds too academic, many busi-ness people will find it alienating. This mustn’t be, because many of these visual analysis techniques are quite easy to learn and can beused to do the kinds of data analyses that arecommonly required in business. In an effort toremove some of the academic stigma, Ben and I have been playing around with new names forthe application of information visualization in apractical way to the everyday needs of business,

which is how the term “BizViz” came to be. Bensuggested “BizViz” along with a few othernames, and this is the one that we both agreedto use to get your attention and tempt you to trysomething new that we know is good for you.

In this paper I’ll do my best to introduce you—people who must make sense of information tosolve real business problems—to a few simpledata analysis techniques that you can use todiscover meaning in your data that might otherwise remain invisible using more traditionalapproaches. I’ll use Spotfire DecisionSite, apopular visual analytics software package, to illustrate these techniques. All of these techniques leverage the strength of our mostpowerful sense: vision.

I SeeBecause vision is our dominant sense, the actsof seeing and thinking are intimately connected.It is not an accident that we use expressionslike “I see” to describe the experience of under-standing. In fact, almost every word we use todescribe understanding, including insight, illumination, and enlightenment, are visualmetaphors. Of all the sense receptors in thehuman body, 70% are located in our eyes. Notonly does vision offer a richer, more nuancedperception of the world around us than ourother senses, it does so through a significantlybroader bandwidth at much higher speeds of delivery. Researchers focus primarily ondeveloping visual methods of exploring and representing information because this is thechannel that can deliver the richest perceptionpossible, resulting, when done effectively, in therichest understanding. To fulfill its promise, BIsoftware must incorporate visual analysis methods—not just any visual analysis methods,but those that actually work.

3

Information visualization—technologies thatsupport the analysis and communication ofdata using visual media and techniques—should not be seen as separate from BI. As theGartner Group’s definition of BI made clear,when data visualizations are used to support an“interactive process for exploring and analyzingstructured and domain-specific information todiscern trends or patterns”, they are doing precisely what BI is meant to do. When usedeffectively, visualization software extends thereach of traditional BI to new realms of under-standing—not as one means among many, but often as the only effective means available.Information visualization will enable the nextleap in BI’s evolution.

All BI software vendors have recognized theappeal to their customers of graphical data displays, but few understand their value; fewunderstand what works and what doesn’t, letalone why. Nothing illustrates this more vividlythan the current popularity of dashboards. Likea feeding frenzy among sharks, vendors sensedthe potential market of dashboards and surgedin with bloodlust to devour as much as possi-ble, as quickly as possible, without taking thetime to understand it. Rather than steppingback and asking why these single-screen consolidations of information for rapid monitor-ing of what’s going on were so appealing, theyassessed the situation superficially and beganto compete with one another in an absurd display of flash and dazzle. Vendor after vendorrushed to awkwardly clomp down the fashionrunway to show off their flashy meters, gauges,and traffic lights while buyers sat mesmerizedby the gaudy spectacle. This will change in timeas this market matures and a critical mass ofunhappy customers rebels against these superficial dashboard displays that fail to communicate. Dashboards, like other visualmedia for communication of information, havetremendous potential, but only when properlydesigned to connect and interact with our eyes(how we see) and our brains (how we think).

To harness the power of visual analysis andcommunication, BI vendors must take the timeto do two things that they often overlook:

1. Explore and think about the useful humanends that their technology should supportuntil they clearly, accurately, and fully under-stand them.

2. Study the research findings that will teachthem how to address these human endsthrough the use of technology in ways thatactually work.

In the case of dashboards, this involves focuson the communication ends that are the exclusive purpose of dashboards, along with astudy of the rich research literature thatdescribes how visual perception works and howinformation displays must be designed to takeadvantage of this powerful and efficient channelof communication. BI vendors need to dig themselves out from under the overwhelmingdemands for features and functions that keepthem frenzied trying to bolt things onto theirproducts as quickly as possible. They must lifttheir heads above the fray long enough toremember that they are supposed to be theexperts who bring a clear and commandingvision to the design of BI products. Customersare responsible for describing their needs—thebusiness ends that they need to achieve.Software vendors are responsible for expertlydesigning technology that supports these needsin ways that really work. Given the central goalof BI to help business people make sense ofinformation to enable smart decisions, BI vendors need to be experts in the visual tech-niques that alone can enable much of the datadiscovery and analysis that businesses require.

If you’re one of those folks who consider them-selves more verbal than visual, and are thereforeuninterested in visual analysis because it doesn’tspeak your language, I’ll let you in on an impor-tant truth: despite your personal preference forverbal over visual information displays, some ofthe messages that are contained in data are visual by their very nature, and therefore remain

4

hidden or difficult to see at best when presentedverbally (that is, as text, such as in a tabular display). I’ll illustrate this point very simply. Take a look at the following table (Figure 1) of time-series sales data for two regions: domestic andinternational.

This table does a marvelous job of giving youprecise values and makes it easy to look up aparticular value or even compare individual val-ues to one another, but there’s a bigger picturethat isn’t obvious in this tabular display. I usethe word “picture” to describe this message inthe data, because it is hard to discern except inthe form of a picture.

Now look at the same exact data (Figure 2) displayed graphically. Previously veiled aspects of the data now pop out clearly and immediately,recognizable and understandable without effort.

Here are a few of the facts that are clearly visiblein this graph:

n Domestic sales are trending upwards overallthrough the year.

n International sales hold reasonably steadythroughout the year, except in the month ofAugust (vacation month in Europe), when theydip.

n Domestic sales fluctuate quite a bit, and witha moment’s attention, we see that they fluctu-ate in a pattern that is cyclical per quarter.They drop in the first month of each quarterand then steadily rise to a dramatic peak inthe last month of each quarter.

Even though each of these patterns could havebeen discerned from the table given enough timeand attention, they certainly could not have beendiscerned so quickly and easily. Even if you favortext displays, being verbally oriented like I alsotend to be, you can’t afford to ignore visual representations when looking for messages thatare contained in the shape of the data.

Even the federal government has recognized thepotential of visual analytics. A new governmentprogram is establishing regional research centersat universities throughout the country to developvisual techniques and tools for exploring andmaking sense of data (see http://nvac.pnl.gov).Funded primarily by the Department ofHomeland Security, but available for participationby all government agencies, these research cen-ters are bringing together academic, commercial,and government talent to leverage the power of visual analytics for greater insight by thosewho make the decisions that affect our nationalinterests. Their motto is “detecting the expect-ed...discovering the unexpected.” I doubt thatthis program would exist if the potential benefitsof visual analytics were not beyond question.

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov DecDomestic 1,983 2,343 2,593 2,283 2,574 2,838 2,382 2,634 2,938 2,739 2,983 3,493International 574 636 673 593 644 679 593 139 599 583 602 690

2005 Sales Revenue(As of 1/1/2006)

Domestic

International

0

500

1,000

1,500

2,000

2,500

3,000

3,500

4,000

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

U.S. $

Figure 1

Figure 2

5

the late 18th and early 19th centuries that manyof the graphs that we use today were invented or dramatically improved by a Scottish social scientist named William Playfair, including barcharts and pie charts. Over a century passed,however, before the value of these techniqueswas sufficiently recognized for the first universitycourse in graphing data to be offered, originally in1913 at Iowa State.

The person who really recognized the power ofvisualization as a means to explore and makesense of data was the Princeton statistics pro-fessor John Tukey. In 1977, Tukey introduced awhole new approach to analyzing data calledexploratory data analysis. Later, in 1983, datavisualization guru Edward Tufte, published his

A Brief History of Information Visualization

Figure 3

To fully appreciate what information visualizationis and what it offers, it’s worthwhile to quicklytrace the historical highlights (Figure 3).

The tabular presentation of data has been withus since the 2nd century, when it was first usedin Egypt to organize astronomical informationand to aid navigation. The representation ofquantitative data in the form of two-dimensionalgraphs, however, didn’t arise until much later, inthe 17th century. Rene Descartes, the Frenchphilosopher and mathematician famous for thewords “Cogito ergo sum” (“I think therefore Iam”), invented 2-D graphs using X and Y axes,not originally for presenting data, but for perform-ing a type of mathematics based on a system ofcoordinates relative to the axes. It wasn’t until

6

groundbreaking book The Visual Display ofQuantitative Information, which showed us thatthere were effective ways of displaying datavisually and then there were the ways that mostof us were doing it, which did not communicatevery well. One year later, in 1984, while watch-ing the Super Bowl, Apple introduced us to thefirst popular and affordable computer thatemphasized graphics as a mode of interactionand display, a graphical user interface (GUI) thatwas originally developed at Xerox PARC (PaloAlto Research Center). This paved the way finally for the use of data visualizations that were interactive.

Given the availability of affordable computerswith reasonably powerful graphics, a newresearch specialty emerged in the academicworld, which was coined “information visualiza-tion.” In 1999, the book Readings in InformationVisualization: Using Vision to Think, collectedthe best of this work into a single volume andmade it accessible for the first time beyond thewalls of academia. This book was co-authoredby Ben Shneiderman, Stuart Card, and JockMackinlay. In it they told us what they meant bythe term information visualization by providingthe following definition:

Information visualization is the use of computer-support interactive visual representations ofabstract data to amplify cognition.

This definition places information visualizationsmack dab in the middle of business intelli-gence. It’s worthwhile to break it apart to highlight and clarify each of its components:

n Computer-supported: The visualization isdisplayed by a computer, usually on acomputer screen.

n Interactive: The visualization can bemanipulated directly and simply by the userin a free-flowing manner, including suchactions as filtering the data and drilling downinto details.

n Visual representations: The information isdisplayed in visual form using attributes likethe location, length, shape, color, and size ofobjects to form a picture of the data andthereby reveal patterns, trends, and excep-tions that might not be seen otherwise.

n Abstract data: Information such asquantities, processes, and relationships, asopposed to visual representations of physicalobjects, such as geography (that is, a map) orthe human body (for example, an MRI image).

n Amplify cognition: These visualizations andthe interactions they enable extend our abilityto think by assisting memory and represent-ing the information in ways that our brainscan easily comprehend.

This is precisely what we need.

Information visualization is the use of computer-support interactive visualrepresentations of abstractdata to amplify cognition.

7

Information visualization research has produced awealth of useful techniques for discovery andanalysis, but a few stand out as especially simpleand effective means to analyze quantitative business data. I’m going to describe and demon-strate four techniques that you can use today:

1. Filtering data directly and instantly2. Extending your view to more dimensions

using multiple comparative graphs3. Viewing data from multiple complementary

perspectives simultaneously4. Highlighting subsets of data simultaneously

and automatically in multiple views

Filtering Data Directly and Instantly

One of the most common steps that you takeover and over again when examining and tryingto make sense of data is that of filtering.Whenever you see something interesting, youneed to take a closer look at it without the distraction of other data, so you filter out thedata that isn’t relevant to your immediate interest. This is part and parcel of the dataanalysis process—nothing new—but the waythat filtering is done with most traditional BIsoftware is clumsy and tends to interrupt thefree flow of analysis.

Here’s how this process typically goes using traditional software: (1) you examine your datain a graph and notice something interesting, (2)you turn away from the data to construct aquery to filter the data and then you run thequery, (3) you wait for the query to return the filtered results, (4) you construct a new graph to view the filtered data, and (5) you turn backto the data to examine it in its new form. Thisfragmented process introduces an annoyingstutter into your thinking process and can actas a disincentive to inquiry and exploration. Itslows you down and sometimes causes you tolose your train of thought.

Wouldn’t it be helpful if you could filter the datain a way that didn’t require you to take youreyes off of it and allowed you to see the effectsof the filtering process as it’s happening, withoutdelay? Information visualization researchers recognized this need several years ago andhave worked hard to provide solutions, whichthey call dynamic queries.

Let’s look at this in the context of a real-worldanalysis of business data. Imagine that we areresponsible for making sense of our company’ssales. We sell five different wines to retailers andhave a particular interest in the sales of Merlot.We have the data to examine sales of Merlotacross two years compared to overall winesales, distributed across four regions (WE, MW,NE, and SW), thirty territories (1-30), and threecustomer types (large, medium, and small), toname a few of the available items. (Note: Thedata that we’ll be using is not real, but was created for the purpose of demonstration only.)Now, let’s focus on Merlot’s percentage of overall sales in each of the two years across thefour regions by displaying it as a simple bargraph (Figure 4).

Simple Information Visualization Techniques forBusiness Intelligence

Figure 4

8

It only takes a second to notice that Merlot’sshare of overall sales has decreased from year1 to 2 in all but region WE (west). This leads usto wonder if this pattern holds true for all typesof customers—big, medium, and small. Noticethat to the right of the graph there is a panellabeled “Query Devices”. This panel includessimple mechanisms for filtering the data, one foreach column that is available (Customer ID,Customer Name, Customer Size, etc.). Let’sfocus for now on the query device for theCustomer Size column. Because Customer Sizeis a categorical variable—one that consists of alimited set of discrete rather than continuousvalues—the query device is in the form of acheck box for each of the values. Right now,each value is checked, but if we want to quicklysee Merlot’s share of sales for big customersonly, we can easily uncheck medium and small[highlighted in red below (Figure 5)] to view thefollowing result:

The general pattern holds true for big customers,but the decrease of Merlot’s share in the SE(southeast) region is not as pronounced. Whatyou couldn’t see, because you’re reading myexplanation of this process rather than doing ityourself and seeing what happens, is that therewas no delay between making filtering selectionsand seeing the results in the graph. For thissame reason you also might not be able to fullyappreciate the fact that this act of data filtering

was very convenient. There was no disconnec-tion between wondering what the pattern mightlook like for big customers alone, filtering thedata, and seeing the result. It involved a fluidinteraction of thought and action.

Let’s pursue this line of investigation further byfiltering out all but the west region to see if allstates in this region display a similar pattern.Because we’re focusing on the change inMerlot’s share from year 1 to year 2, we’llchange the graph to show a single bar perstate that measures the amount of change,rather than a separate bar for each year. Here’swhat the data looks like now (Figure 6).

Of the four states in the west region, Idahostands out as an exception to the pattern of anincrease in Merlot’s share. Now let’s use thequery devices to do something interesting. Oneof the measures that is available in our data isthe number of visits that were made by oursales team to each of the customers in the second year. Let’s see if there is a relationshipbetween the number of visits and the fact thatMerlot’s share decreased in Idaho. The querydevice for the number of visits is displayed inthe form of the slider, shown below (Figure 7):

Figure 5

Figure 6

Figure 7

9

This type of slider, with a control at each end,allows you to easily set the range of values thatyou want to see to any range you want. Noticethat the low and high ends of the slider arelabeled with the values 0 and 26, which indi-cates the entire range of values in the numberof visits measure. To filter the data based on thenumber of visits, in most BI software you wouldhave to define the range, submit the query, andwait for the results, but with this slider, we can adjust the range dynamically andsee the results change in the graph as we doso. For instance, we can adjust the slider toinclude only those customers who received oneor more visits, then two or more visits, and soon, seeing the results of each adjustment in thegraph as we proceed. Nothing much happenswhen we adjust the slider from one or more visits through three or more visits, but look atthe change that occurs below from left to rightwhen we adjust the slider from four or more visits to six or more visits (Figure 8):

With five or more visits (middle graph), thechange in Merlot’s share increased a little inCalifornia and Oregon, but with six or more visits(right graph), Idaho disappeared altogether. Inother words, none of the customers in Idahoreceived more than five visits from the sales team out of a maximum of 26 visits that some customers received. Being able to see the datachange in the graph as we’re filtering it causedthis meaningful aspect of the data to leap out.

Dynamic queries, enabled through simple filtering controls like sliders, offer an enormous

Figure 8

advantage when exploring data. It is so easy todo, you needn’t take the time to construct fullyformulated queries before pursuing a line ofinquiry, but can approach the process moreimprovisationally, as an effortless series ofthoughts, actions, and observations.

Extending Your View to MoreDimensions Using MultipleComparative Graphs

You might have noticed how helpful it was tosee multiple instances of the same graph all atonce in Figure 8 below, each representing a different value of the number of sales visits vari-able. This technique of arranging a consistentseries of the same graph, differing only along asingle variable (in this case, number of visits), allwithin eye span so they can be compared toone another is a powerful way to add anotherdimension to the data that would be difficult todisplay in a single graph. In 1983, data visuali-zation guru Edward Tufte called this technique“small multiples”, but it goes by other names aswell, including a trellis display, which is whatSpotfire calls it. When multiple graphs arearranged in this way, especially as a full matrixincluding multiple columns and rows of graphs,it looks a bit like a trellis that might be found inyour garden. Very few software vendors make iteasy to arrange graphs in this manner, despitethe value of this technique for analysis and thefact that it has been around for over 20 years.

You can use this kind of display to see a greatdeal of data in a way that causes interesting facts

10

to jump out that might otherwise remain hidden.Let me illustrate using an example that includes50 graphs—one for each state—as well as theamount of change in total sales (black bars) andMerlot sales (gray bars) separately for big, medium, and small customers (Figure 9).

Obviously, a display like this isn’t designed forprecise comparisons between individual values,but rather for seeing the big picture such thatanomalies can surface. My eyes are drawn to afew interesting facts in this display:

1. Overall wine sales to small customers inDelaware increased dramatically, but Merlotsales decreased.

2. Merlot sales in New Hampshire decreaseddramatically among big customers butincreased by about the same degree amongmedium-sized customers.

3. Overall wine sales took a dive in Mississippiamong big customers, yet Merlot sales stillmanaged to increase a bit.

4. We didn’t sell any wine whatsoever to bigcustomers in Montana.

5. In Washington, Merlot sales increased signifi-cantly among customers of all sizes, and didso to a greater degree than overall wine sales.

More can be found with a little study, but this isenough to make the point: interesting data canbe rapidly detected using a series of relatedgraphs that can all be seen within eye span thatmight never be found if you were forced toexamine the same data one graph at a time.

Viewing Data from MultipleComplementary PerspectivesSimultaneously

Multiple graphs can be displayed together inanother way that is also quite powerful foranalysis. Entirely different views of the samedata, with each table or graph showing the datafrom a different perspective, can be quiterevealing when they are seen together.

One of our most fundamental limitations whenanalyzing data resides in short-term memory.This is that part of memory that is active whenwe are consciously thinking about something. Nomore than four to eight chunks of informationcan be held in short-term memory at any onetime. Traditional BI methods of analysis are oftenso fragmented that significant observations havealready been forgotten by the time we make ournext observations, which means that we cannotcompare them to one another. Most of the

insights that we gain through analysiscome in the form of comparisons.Consequently, good data analysis softwareneeds to compensate for the limits ofshort-term memory, primarily in two ways:

1. By organizing meaningful data into larg-er chunks so that the four to eight chunksof information that we can hold in short-term memory at any one time are as richas possible.2. By placing as much of the data that weneed to compare within eye span so wedon’t need to keep it all in short-termmemory to make the comparisons.

Figure 9

11

Remember the table of numbers in Figure 1?Each of the 24 numbers represents a chunk ofinformation when held in short-term memory,which means that only a handful of them canbe considered at any one time. The graph ofthe same data in Figure 2, however, representsthe values for each category (domestic andinternational sales) as a single line in thegraph—a visual object that we can hold inshort-term memory as a single chunk of information. This is one of the reasons thatgraphs are such a powerful medium for theanalysis of data.

When comparisons need to be made betweendifferent values, sets of values, or views of thedata, they need to be available to our eyes at thesame time. Nothing needs to be held for long inshort-term memory when it is all there before oureyes and we can quickly swap the chunks thatwe need in and out of memory as we’re makingconnections and considering what they mean.Software can support this need by allowing us toplace several different data displays on a singlescreen. Let’s look at a rich assortment of per-spectives on the same set of sales data thatwe’ve been exploring (Figure 10).

Starting with the graph in the upper left-handcorner and working down each columnbefore moving right to the next, here’swhat we see:

1. Total wines sales (black bars) com-pared to Merlot sales (gray bars) inU.S. dollars for this year per region.

2. Percentage change in sales betweenlast year and this year—total andMerlot—per region.

3. Merlot’s percentage of total winesales per region.

4. Change in Merlot’s percentage oftotal sales from last year to this yearper region.

5. Total wine sales this year per customer size(darkest gray represents big customers,medium gray represents medium-sized cus-tomers, and light gray represents small cus-tomers) by region.

6. Merlot sales this year per customer size byregion.

7. Merlot share of total sales this year percustomer size by region.

8. Frequency distribution of sales based on thenumber of sales visits that led to each sale(0-3, 4-7, 8-11, etc.).

9. Scatter plot showing the correlation of thepercentage change in total wine sales fromlast year to this year (vertical axis) to thepercentage change in Merlot wine sales fromlast year to this year (horizontal axis).

Each of these graphs provides a different perspective on our sales. Seen together, theyhelp us see connections (interesting interac-tions) between various aspects of sales. Studythese graphs for a few minutes to see whatworthwhile stories they tell you about sales.Here are a few thought-provoking features ofthe data that caught my attention while examining these graphs:

Figure 10

12

1. Merlot sales in the west region representaround 55% of total sales, the greatest shareof any region, but this is mostly due to adramatic increase in Merlot sales in the west.Despite this strong showing in the west,Merlot’s share of total wine sales actuallywent down a little in the midwest andnortheast and increased only slightly in thesoutheast. Merlot’s high share of totalsales in the west is strongest among smallcustomers.

2. Even though the west barely beat thenortheast in overall wine sales this year tocome in third place of the four regions, theycame in a strong second in Merlot sales,nearly catching up with southeast. In fact,Merlot sales to medium-sized customers inthe west roughly match those in the south-east.

3. Most sales resulted from between three andseven visits to a customer and the numberof visits that seems to be required for salesis fairly consistent between big, medium, andsmall customers.

4. While there is a positive correlation betweenchanges in total wines sales vs. Merlot salesfrom last year to this year (that is, as totalwine sales increased or decreased, Merlotsales tended to do the same), there are anumber of exceptions to this pattern,especially in the lower right corner where adisproportionate number of Merlot salesincreases appear relative to decreases inoverall sales.

Many of these discoveries, as well as othersyou perhaps made while examining this collection of complementary graphs, might have remained invisible using more traditional BImethods of analysis, and certainly would havebeen much more difficult and time-consumingto ferret out.

Highlighting Subsets of DataSimultaneously andAutomatically in Multiple Views

This next technique builds on top of the one wejust examined. The revelatory power of multipleperspectives on the data seen together can beextended through a technique that informationvisualization researchers call brushing. Let’s saythat we are looking at one of the graphs inFigure 10 and we become interested in a particular subset of the data, such as thoseMerlot sales that decreased since last year eventhough overall wines sales increased, as shownin the upper left quadrant of the scatter plot.Now imagine that we have a brush that we canuse to paint a rectangle around these particulardata points in the scatter plot to highlight them,resulting in the following (Figure 11):

Now all the values of overall wine sales that aregreater than zero corresponding to Merlot salesthat are less than zero are highlighted in red.Simply making them stand out in the scatter plot, however, is not the point of this exercise.What we really want to see is how this particularsubset of data behaves in all of the other graphson the screen. That’s exactly what brushing doesfor us, and it does it automatically. Take a looknow at the full screen and see if it leads you toany interesting insights (Figure 12).

Figure 11

13

One of the first things that I noticed is that thesedecreases in Merlot sales that are out of syncwith corresponding increases in overall winesales, occurred less often in the west (especiallyseen in upper graph in the left column and themiddle graph in the center column). Relative tocustomer size, this pattern also seems to be disproportionately strong in the southeast (seethe top and middle graphs in the center column).

Now let’s say that we want to see if sales in thewest fall disproportionately in any particular area of the scatter plot. To see this, we canbrush the west region in any of the graphs inthe left or center columns and see the resultshighlighted in every graph, including the scatterplot shown below (Figure 13).

Notice that a disproportionate number of salesin the west (highlighted in red) appear below thetrend line and in the right half (increases inMerlot sales) of the graph. This reaffirms ourprevious observation that the west has con-tributed more than other regions to increases inMerlot sales, especially when overall wine saleschanged less than the average degree.

Final WordEven with this simple data set, we could go onfor hours pursuing lines of investigation asquickly as they come to mind. When the kindsof analytical functionality that we’ve just exam-ined are enabled by software in such simple andefficient ways, the step-by-step process thatbegins with observation, then raises a question,

and is followed by manip-ulation of the data to pur-sue that question, result-ing in new observationsand insights, becomesfluid, without interruption.When this happens, wecan achieve a state ofawareness and insight thatpsychologist MihalyCsikszentmilhalyi (pro-nounced “chick-sent-me-high”) calls “flow” and oth-ers more colloquially callbeing “in the zone.” Thiscan only happen when ourawareness of the softwarewe are using recedes into

the background and we become fully immersedin the data and its rich story.

With the right tools you can clear the fog andlearn to analyze data at the speed of thought.Take advantage of the techniques that informa-tion visualization researchers have developed tohelp you work smarter. Achieve the promise ofbusiness intelligence today.

Figure 13

Figure 12

Spotfire® DecisionSite®

Spotfire, Inc.212 Elm StreetSomerville, MA 02144 U.S.A.Telephone +1.617.702.1600Fax +1.617.702.1700Toll-Free +1.800.245.4211

Spotfire AB(European Headquarters)Första Lânggatan 26SE-413 28 Göteborg, SwedenTelephone +46.31.704.1500Fax +46.31.704.1501

Spotfire Japan KK(Japanese Headquarters)Kinokuniya Bldg. 7F, 13-5,Hatchobon 4-chromeChuo-Ku, Tokyo 104-0032 JapanTelephone +81.3.5540.7321Fax +81.3.3552.3166

www.spotfire.com

About Spotfire

For thousands of business users faced with day-to-day decisions, Spotfire analytics offers theplatinum user experience for visually interactingwith information. Distinguished by its intuitiveease and analytic power, Spotfire software rapidlyreveals unseen threats and illuminates newopportunities, creating unprecedented economicvalue. Spotfire’s customers include industry leaders among the Global 2000 that havedeployed Spotfire analytics to gain an informationadvantage over their competitors. For more information, visit http://www.spotfire.com.