Predictive analytics in action real-world examples and advice
Predictive Analyticsin Action: Real-WorldExamples and AdvicePredictive analytics projects are inherently complex and potentiallycostly. But for organizations that get it right, they can pay off in improveddecision making and competitive advantages over business rivals.An Essential Guide1 2 3 4Editor’s Note PredictiveAnalyticsProgramsNeed OpenOrganizationalMindsRecipefor PredictiveAnalyticsSuccessIncludesOne PartStorytellerSurveys Pointto Skills, Trainingas PredictiveAnalyticsHurdlesVirtualizationCloudApplicationDevelopmentNetworkingStorageArchitectureDataCenterManagementBusinessIntellegence/ApplicationsDisasterRecovery/ComplianceSecurity
2 Predictive Analytics in Action: Real-World Examples and AdviceHomeEditor’s NotePredictiveAnalytics ProgramsNeed OpenOrganizationalMindsRecipe forPredictive AnalyticsSuccess IncludesOne PartStorytellerSurveys Point toSkills, Training asPredictive AnalyticsHurdles1editor’s noteModeling the Future:A Challenging but RewardingPropositionWho wouldn’t want to predict the future, especially when money is atstake? Alas, businesses can’t just rely on crystal balls, tarot cards and palmreaders—at least if they want to stay in business. But companies can turn topredictive analytics software to help them peer into the business future—forexample, to predict which customers are likely to be open to cross-selling of-fers and which ones might not be worth additional sales attention.But they don’t call it advanced analytics for nothing. If your organization islooking to deploy and use predictive analytics tools, you’ll need to make surethat you have the right level of analytics skills in place. Time for an infusionof data scientists, perhaps? Building predictive models is a complex, time-consuming process that requires trial-and-error testing in order to get thealgorithms to produce the desired analytical results. And convincing busi-ness and operational managers to trust what the models are telling them andadjust their strategies and processes accordingly is another big challenge.This three-part guide offers practical advice from experienced analyticsprofessionals and industry consultants on how to successfully manage a pre-dictive analytics program. The lead story details key steps to take in develop-ing and implementing a program, starting with ensuring that your companyis open to the possibilities enabled by predictive analytics. Next, we recountthe lessons that one analytics exec has learned about building a predictiveanalytics team. And we report on a pair of surveys pointing to a lack of skillsand proper training as predictive analytics inhibitors.Craig StedmanExecutive Editor, SearchBusinessAnalytics.com
3 Predictive Analytics in Action: Real-World Examples and AdviceHomeEditor’s NotePredictiveAnalytics ProgramsNeed OpenOrganizationalMindsRecipe forPredictive AnalyticsSuccess IncludesOne PartStorytellerSurveys Point toSkills, Training asPredictive AnalyticsHurdles2BestPracticesPredictive Analytics ProgramsNeed Open Organizational MindsHas the current fervor to pounce on every piece of available data for po-tential analytical uses spawned a world in which information often is col-lected for its own sake? Sometimes it might seem that way. But in theever-expanding universe of “big data,” predictive analytics software is onetechnology that can take advantage of the great variety of data accumulatedby an organization as it works to model customer behavior and future busi-ness scenarios.And using predictive analytics tools to interpret data is becoming moreimportant to businesses: The most successful companies and rising-star en-terprises sedulously employ them to help point the way forward on businessstrategies and operations, according to analysts who focus on advanced ana-lytics technologies. But that doesn’t happen magically, they cautioned; orga-nizations need to take the right steps to develop effective predictive analyticsprograms.In many industries, getting a leg up on the competition can be more chal-lenging than ever—especially if companies are set in their ways. The startingpoint in embracing predictive analytics should be ensuring that an organiza-tion has a proper frame of mind about using the technology, the analysts said.An open, dexterous attitude that’s naturally curious, eager to learn and will-ing to adapt will produce the best results.Douglas Laney, an analyst at Gartner Inc. in Stamford, Conn., thinks a pre-dictive analytics program should begin by questioning historical businessmethods while searching far and wide for better ones. Companies “shouldnot only focus on how things have been done in the past but be open tobig ideas for innovations and transformations,” he said.“This could mean
4 Predictive Analytics in Action: Real-World Examples and AdviceHomeEditor’s NotePredictiveAnalytics ProgramsNeed OpenOrganizationalMindsRecipe forPredictive AnalyticsSuccess IncludesOne PartStorytellerSurveys Point toSkills, Training asPredictive AnalyticsHurdles2BestPracticesapplying measures effective in other industries to your industry.” Such amind-set should extend to the point of embracing approaches that “radicallychange the way business processes are done” in an organization, Laney added.With that in mind, the mentality of the players—particularly the businessmanagers who are being asked to buyinto the findings of predictive mod-els—is frequently the key variable thatdetermines the success or failure ofpredictive analytics programs. A per-spicacious corporate culture championsobjectivity, welcomes new ideas and isnaturally flexible. Conversely, a retro-grade one resists change and draws heavily on existing biases and subjectiveformulas.“Resisting new ways of doing things is the reason most projectsfail,” said John Lucker, head of Deloitte Consulting LLP’s advanced analyticsand modeling practice.Keep Your Eyes on the Business PrizeThe grand plan of a predictive analytics deployment should also begin with aclear set of business objectives, said Thomas “Tony” Rathburn, a senior con-sultant at The Modeling Agency LLC, a Pittsburgh-based consulting com-pany that focuses on data mining and predictive analytics. Then, he added,a team-oriented strategy is needed to advance those objectives. That is bestconstructed through substantive discussions involving program managers,predictive modelers, data analysts and business representatives.So critical is the strategy development process that Eric King, presidentand founder of The Modeling Agency, recommends retaining “a seasonedstrategic mentor” to help lead the effort and keep it on track.Once a predictive analytics strategy is in place, it’s time to begin the anal-ysis process. Laney said “chewy” questions that probe deeply into data willunearth findings with high operational value. The truly useful ones, he said,are multifaceted—for example,“How can we grow new customers by 20%Once a predictiveanalytics strategyis in place, it’stime to begin theanalysis process.
5 Predictive Analytics in Action: Real-World Examples and AdviceHomeEditor’s NotePredictiveAnalytics ProgramsNeed OpenOrganizationalMindsRecipe forPredictive AnalyticsSuccess IncludesOne PartStorytellerSurveys Point toSkills, Training asPredictive AnalyticsHurdles2BestPracticesper year for a certain product line without cannibalizing other product linesgiven the range of economic forecasts, competitor trends and changing con-sumer demands?” Run through predictive models, such questions can con-tribute in a big way to driving new business, according to Laney.Building Models is a Testing ProcessAfter choosing and deploying the predictive analytics tools that best fit thejob at hand, developing models is the next step. Mike Gualtieri, an analyst atForrester Research Inc. in Cambridge, Mass., said analytics algorithms shouldbe run on 70% of a data set to create an effective predictive model.“Thenyou test that model on the remaining30%,” he said.Completed models should be regu-larly tested and enhanced as needed,and a set of performance metricsshould be put in place for tracking theiraccuracy, Gualtieri added—all part of aprocess for “continuous monitoring ofthe predictive analytics model.”Moreover, said other analysts, theentire predictive analytics process requires regular monitoring as businessneeds and the nature of the data being collected by an organization change.Analytics strategies and tactics that worked initially will need to be revisitedand perhaps revised in order to continue achieving optimal results.The mark of a truly successful predictive analytics program, Lucker said,is when some of the cost savings or business gains realized from an ongoinganalysis project can be applied to pay for the next one so no new dollars needto be spent.“Using the value of each project to fund downstream efforts is anevolutionary approach that comes with a [built-in] return on investment,” hesaid. —Roger du MarsAnalytics strategiesand tactics that workedinitially will need tobe revisited and per-haps revised in orderto continue achievingoptimal results.
6 Predictive Analytics in Action: Real-World Examples and AdviceHomeEditor’s NotePredictiveAnalytics ProgramsNeed OpenOrganizationalMindsRecipe forPredictive AnalyticsSuccess IncludesOne PartStorytellerSurveys Point toSkills, Training asPredictive AnalyticsHurdles3TeamBuildingRecipe for PredictiveAnalytics Success IncludesOne Part StorytellerThe secret to building a successful predictive analytics team is findingpeople with statistical analysis, programming and—perhaps most impor-tant—storytelling skills, according to one practitioner.It’s important to find multitalented people because, oftentimes, predictiveanalytics teams are rather small, said Jennifer Golec, vice president of strate-gic analytics at XL Insurance Inc. Multifaceted individuals offer a higher levelof flexibility, she said, and that comes in handy when resources are tight.Ideally, predictive analytics professionals should be one part programmer,Golec said, because they’ll be working with a great deal of information andconducting exploratory analysis. Commercial software can help in these ar-eas, she explained, but some programming skills will still be helpful for taskslike manipulating or massaging dataand creating new variables.Predictive analytics professionalsshould also focus on developing sta-tistical analysis skills because thoseare necessary for building multivariatemodels, statistical tools that use multi-ple variables to forecast outcomes.“The third piece is that you have to be part storyteller. You have to be ableto interpret those results,” Golec said.“[That means] really being able to in-terpret the insight that you pull from the data. You have to be able to relaythat because if you don’t, you’ll be sitting on this great model and you won’tbe able to implement it.”Predictive analyt-ics professionalsshould focus ondeveloping statis-tical analysis skills.
7 Predictive Analytics in Action: Real-World Examples and AdviceHomeEditor’s NotePredictiveAnalytics ProgramsNeed OpenOrganizationalMindsRecipe forPredictive AnalyticsSuccess IncludesOne PartStorytellerSurveys Point toSkills, Training asPredictive AnalyticsHurdles3TeamBuildingMore Than Just Crunching NumbersThe popular 2011 film Moneyball—which tells the story of Oakland A’s gen-eral manager Billy Beane, who used analytics to find undervalued players andbuild a great baseball team—might give the mistaken impression that ana-lytics is all about crunching numbers. But it’s much more than that, accord-ing to Golec. Organizations must also strive to understand how the results ofpredictive models translate to the real business world.“Sometimes that is the danger withproducts like SAS,” Golec said.“Theymake it so easy to push the data in andhit the button and have somethingcome out. But if you’re not trained tounderstand and interpret that output,you could end up with junk and youmight not know it.”Golec, who has a doctorate in eco-nomics from the University of Missouri and previously ran a predictive ana-lytics program for insurance provider The Hartford, began working for XL Insurance and its global parent company, XL Group PLC, in October 2011.Analytics Goal: Ratcheting Back on RiskOne of her first tasks was to find a software vendor that could help the prop-erty and casualty company build out its fledgling predictive analytics pro-gram. XL Insurance launched the program to do a better job of avoidingunnecessary risk and, ultimately, improving its loss ratio.“The loss ratio islosses over premiums,” Golec said.“The lower it is, the more profitable youare.”Golec took a close look at SPSS, which was acquired by IBM in 2011, andWolfram Research’s Mathematica tools. But she had worked with softwarefrom SAS Institute Inc. in the past and decided to do so again.“Half the battle is working with the data, manipulating the data and get-ting it into a form that allows you to actually do the modeling,” she said.Organizations mustalso strive to under-stand how the resultsof predictive modelstranslate to the realbusiness world.
8 Predictive Analytics in Action: Real-World Examples and AdviceHomeEditor’s NotePredictiveAnalytics ProgramsNeed OpenOrganizationalMindsRecipe forPredictive AnalyticsSuccess IncludesOne PartStorytellerSurveys Point toSkills, Training asPredictive AnalyticsHurdles3TeamBuilding“SAS allows us to get the data into the shape and form that we want.”XL Insurance is using several SAS products, including SAS/STAT, a statis-tical analysis tool; SAS Graph, a visual tool that allows users to present in-formation in charts and graphs; SAS Enterprise Guide (EG), which makes iteasier to do exploratory analysis of data stores; and JMP, a data visualizationtool.Ensuring Adoption Central to Implementation ProcessThe team at XL Insurance is in the process of building predictive models forrisk assessment. The next step, according to Golec, is to implement thosemodels and closely monitor and measure the results.Golec said the toughest aspects of the implementation phase will likely re-volve around change management and, specifically, getting the right peopleto adopt predictive analytics findings as part of their usual routines. Makingsure that any workflow or architecture changes are properly documented isalso a major challenge.Another is “making sure that we’ve come up with how we’re going to track it and make sure it’s working,” she said.“But I think the big thing in implementation is just achieving that buy-in and making sure that it’s used.”—Mark Brunelli
9 Predictive Analytics in Action: Real-World Examples and AdviceHomeEditor’s NotePredictiveAnalytics ProgramsNeed OpenOrganizationalMindsRecipe forPredictive AnalyticsSuccess IncludesOne PartStorytellerSurveys Point toSkills, Training asPredictive AnalyticsHurdles4ChallengesSurveys Point to Skills, Trainingas Predictive Analytics HurdlesBusinesses recognize the potential of predictive analytics, yet there’sa large gap between those who see it as important and those who actu-ally use the technology, according to a pair of surveys conducted by VentanaResearch.The market research and consulting company, based in San Ramon, Calif.,conducted an initial study in early 2011 which found that only 13% of the re-sponding organizations were using predictive analytics. But 80% indicatedit was important or very important, said David Menninger, who was infor-mation technology research director at Ventana when he was interviewed forthis story; he has since taken a job with a technology vendor.The reason for that gap? While most businesses consider predictive analyt-ics important, those that struggle with it lack both the skills and the trainingrequired to be successful with the technology, Menninger discovered in a fol-low-up study.“Organizations are least mature in the people aspect,”he said.That conclusion was drawn from the results of a three-month survey of198 respondents measured against Ventana’s predictive analytics maturitymodel, which was used to rate the survey responses across the categories ofprocess, information, technology and people.The survey revealed that self-service predictive analytics, or end users cre-ating and deploying their own analyses, has not been widely deployed, de-spite a wave of easier-to-use predictive analytics tools coming to market.Analytics Skills Not a Common TraitIn fact, almost half of the respondents questioned whether users have thebackground to produce their own analyses. For the nonbelievers, Menninger
10 Predictive Analytics in Action: Real-World Examples and AdviceHomeEditor’s NotePredictiveAnalytics ProgramsNeed OpenOrganizationalMindsRecipe forPredictive AnalyticsSuccess IncludesOne PartStorytellerSurveys Point toSkills, Training asPredictive AnalyticsHurdles4Challengessaid it came down to two reasons: 83% reported users didn’t have enoughskills, and 58% reported users didn’t understand the mathematics involved.“[Predictive analytics] requires the specialist skill set—the data scientist,the statistician, the data mining experts—to be successful,” he said.Instead of relying on users, 63% of respondents reported their organiza-tion had a specialized team for predictive analytics or that the task fell tothe business intelligence (BI) and data warehousing (DW) team. But eventhen, Menninger’s research indicated that how satisfied respondents are withthe way predictive analytics is used in their organizations (two-thirds saidthey’re satisfied) depends, in part, onwho does the work.The highest levels of satisfaction,70%, came from respondents whoworked for organizations that employedspecialists such as data scientists toproduce the predictive analytics find-ings. The lowest levels of satisfaction,59%, came from respondents whose BI and DW teams were in charge of the work.“I think it’s common for organizations to think this will naturallyfall out of the BI and DW team,” Menninger said.“But what this tells me isthat this is not a generalized skill of BI and DW teams.”Support Lacking for Predictive Analytics UsersMany organizations are also not doing a great job providing the ongoing sup-port needed to successfully maintain a strong predictive analytics program,Menninger said.According to the survey results, businesses are most successful at provid-ing concept and technique training (44% of respondents felt this was ade-quate) and have the most trouble delivering help desk support (24% reportedthis was adequate). More than a third of respondents, 42%, also found prod-uct training to be adequate.Many organizationsare not doing a greatjob providing the ongo-ing support needed tosuccessfully maintaina strong predictiveanalytics program.
11 Predictive Analytics in Action: Real-World Examples and AdviceHomeEditor’s NotePredictiveAnalytics ProgramsNeed OpenOrganizationalMindsRecipe forPredictive AnalyticsSuccess IncludesOne PartStorytellerSurveys Point toSkills, Training asPredictive AnalyticsHurdles4ChallengesMenninger said concept, technique and product training may drive a stron-ger sense of satisfaction because they require “specialized knowledge” overthe broader needs—and the skills—required by something like a help desk.“I think it relates back to necessary skills,” he said.“How do you have peo-ple on the help desk supporting a morecomplicated topic? The help desk re-sources would need to have specializedtraining and skills to be able to providemeaningful support.”Yet respondents indicated that, inaddition to concept and techniquetraining, the most effective type of support was brought about by help deskresources. Organizations that provided either support feature adequately hadan 89% satisfaction rating on average, according to the survey results.“I suspect that organizations probably think first about doing producttraining and less about this generalized set of skills and help desk resources,”Menninger said.While the level of satisfaction in a predictive analytics program may waxand wane based on training, Menninger said the root of that issue is mostlikely derived from what he considers to be an even bigger problem—a lack ofskills.“The skills issue is significant,” he said.“It appears to have been preventingorganizations in the past from either choosing to tackle predictive analyticsor [being able] to tackle it successfully.”Menninger said predictive analytics requires a deeper kind of knowledge.“It’s unrealistic today to expect the technology to deliver self-service capa-bilities,” he said.“[But] if you have the right skills, the technology is availableto be successful with predictive analytics.”—Nicole LaskowskiThe most effectivetype of support wasbrought about byhelp desk resources.