The continued upward movement of traditional direct marketing combined with the explosive growth in online retail sales creates an unprecedented number of transactions for marketers.  In fact, online retail sales are expected to reach over $360 billion dollars by the year 2016 (Woo, 2012).  This scenario also creates an ever-expanding set of choices for the consumer.  Because of those choices, the astute marketer needs to find a way to fully understand their customers so that they may not only fulfill the customer’s current needs and wants, but anticipate their future needs and wants as well.  The challenge is for businesses to deliver the right offer, at the right time, at the right price, through the right channel (Parr-Rudd, 2012). Anticipating and predicting future customer behavior almost seems like Voodoo, but is in reality child’s play for a kick-ass predictive analytics software package.

     For the uninitiated --Predictive Analytics is a statistical or data mining solution that relies on algorithms, machine learning and various techniques to identify the likelihood of future outcomes based on historical data (Halper, 2014).  Even though that sounds all fancy-schmancy and super-shiny-new-hi-techy, predictive analytics itself is not a new technology by any means.  As a matter of fact, it has been used on structured data for many years—decades even.   It has come into the white-hot spot light more recently though because of the existence of cheaper computing power, an exponential rise in the amount of data available, and the sheer volume of data being collected –also known by the fashionable moniker “big data”— its relative ease of use, and probably most importantly, the increased need for companies to compete more effectively (Halper, 2014).   It is after all, a jungle out there. 

     In the past, marketers relied almost solely on RFM Analytics (which still works just fine, thank you very much) to determine each customer’s valuein time, and to predict how those customers would react to future offers or calls to action based on their historical activity.   RFM is a comparatively simple way for marketers to manipulate their database to determine those customers who should be most responsive to a given marketing effort.  The RFM process uses information about the customer’s Recency, or the date of the customer’s last transaction, Frequency, or the number of transactions made by the customer in a given time period, and Monetary value, which includes the aggregate amount the customer has spent with the organization.  RFM, get it?  Customers are then ranked into 5 quintiles based on their performance in each dimension.  The score of “5” goes to the highest performers and “1” goes to the lowest performers in each dimension.  Once scoring is complete, the three dimensions are realigned in the database to give each customer a total RFM score.  The best customers, and those who are most likely to respond to new offers or calls to action have the highest scores.  Of course a score of 555 is optimal, but not every customer will have that score.  So it becomes the job of the marketer to decide where the reasonable cutoff may be.   RFM scoring and analytics has been a great basis for understanding customer value and predicting future customer activity based on past behavior, but in reality only represents a fraction of the clairvoyance of “Predictive Analytics Software”.

     Predictive Analytics begins with the understanding of a set of predictors, which are really nothing more than single value measurements of each customer (Roebuck, 2012).  Predictors are the foundation upon which predictive analytics is built.  Examples of predictors include the previously mentioned Recency, Frequency, and Monetary Value, but there are also many, many more.   Of course these individual predictors don’t mean much when viewed alone.  Predictive

Analytics actually looks at the relationship between predictors much in the same way that the older, simpler RFM analytics did only with an added mega-dose of steroids.  When predictors are combined and viewed in a relational context, they create models, which in turn allow for inferences to be drawn through Predictive Analytics –also known as Predictive Modeling (Roebuck 2012).

     Predictive analytics “connects data to effective action by drawing reliable conclusions about current conditions and future events” (IBM, 2010).  Of course such magic doesn’t just spontaneously occur.  PA takes people to make it happen, and it all begins with the definition of a business problem or goal.  Marketing and data specialists must collaborate to not only determine if the goal fits current business requirements, but if a relationship can be drawn at all, and if so, will it have any real meaning.  Once the marketer has determined the business problem he wants to solve, he then determines the predictors he feels will best give him the information he is looking for.  PA uses historical data, broken down into individual predictors in order to more fully understand a company’s customers’ actions and to ensure that future marketing activities can be more accurately targeted (Roebuck, 2012).  In order to uncover potential trends and to understand very complex relationships between seemingly disparate variables amongst a sea of data, marketers build models based on two or more individual predictor variables.  The model is then loaded into a sophisticated software program that provides the analysis.     The program is able to extract trends and relationships that an analyst could never uncover in terabytes of data. That information then enables a marketer to place customers into segments that are easier market to as homogeneous groups.  The marketer also has a higher level of certainty as to the potential outcome when targeting groups with marketing initiatives.

     As is the case with all direct marketing, there is an element of testing and tweaking so that

The marketer can adjust the PA results to their most optimal.  We can’t simply take the

computer’s word for it and rollout a very expensive campaign without testing, adjusting and retesting first.  The testing phase confirms and sharpens what the computer and PA software have told us.  Figure 1. illustrates the flow of the process. 

     Since the advent of the “big data” age, and companies’ need to look deeper and deeper into their customer base to uncover every potential dollar in sales, Predictive Analytics has become huge business.  Where there were once just a few companies producing software to cater to those analytical clients’ needs, the market has been flooded with more and more companies looking to get in on the PA trend. Some of the offerings include SAS Predictive Analytics, SPSS (an IBM Company), SAP, Oracle Data Mining (ODM), Predixion, Statistica, RapidMiner, and Angoss just to name a few. It should be noted however that SAS, SPSS, SAP and RapidMiner are the most highly regarded and utilized industry wide (Predictive Analytics Today, n.d.).

     SAS saw its genesis at North Carolina State University where it was originally undertaken as a project seeking to better analyze agricultural research.  As word got out, demand for such powerful analysis software began to grow and the company known as SAS was founded in 1976.  SAS stands for “Statistical Analysis System”, and has become one of the leaders in the

predictive analytics industry (SAS, 2015).  SAS offers enterprise solutions that allow companies

to process massive amounts of data in areas that include Advanced Analytics, Business

Intelligence, Customer Intelligence, Data Management, Decision Management, Risk Management and Supply Chain Intelligence.  These insights are facilitated through a portfolio of SAS products that include, but are not limited to SAS 9.4, SAS/STAT, SAS Analytics Pro, SAS Data Management, SAS Enterprise Miner, SAS Marketing Optimization, SAS University Edition, SAS Visual Analytics, et al. Though the SAS products span a multitude of industries and organization sizes, they will allow marketers of all stripes to gain deeper insights on, and draw inferences from vast amounts of customer data in order to make intelligent marketing decisions moving forward. 

     SAS interfaces nicely with the open-source framework of Hadoop, and if there were any doubt about SAS’s relationship to the platform, they make Hadoop-specific applications as well.  Hadoop was developed by Apache Software as a solution to store and process huge amounts of data and deliver distributed processing power via the cloud for a very low cost.  It allows even the smallest companies to analyze billions of lines of data in just seconds (Mariani & Thompson, 2014).

      SPSS is an IBM predictive analytics software that allows the user to predict with confidence what will happen next so they may make smarter decisions, solve problems, and most importantly, improve outcomes (SPSS, 2015).  The SPSS website has a bevvy of preconfigured

     SAP is a German company doing business in over 130 countries and with over 282,000 customers around the globe (SAP, 2015).  SAP also offers a multitude of software applications for a variety of industries and business types, but does have products that place special emphasis on marketing intelligence.  SAP purports to offer real-time contextual marketing that delivers individualized customer engagements at every stage of the buying process. By employing their software, the marketer would be privy to unprecedented customer insight from marketing analytics and could in turn, leverage that insight into creating a more loyal customer and increasing demand.   SAP is another company that makes mention of using predictive analytics to gain insights into customer attitudes via social media analytics, but this author is less impressed with this feature.  The fact that customer attitudes can change on a dime, and the fact that anything gleaned from social media isn’t exactly valid decision-making data makes this

facet something that may do more to cloud the process than provide any real actionable intelligence.

      Any time there is a massive movement toward a new paradigm (I hate myself for using that word),  there will always be detractors and iconoclasts who try to figure out why the collective wisdom of an industry is wrong.  A recent Forbes article discussed how the notion of customer loyalty is too hard to measure, misleading and difficult to correlate with other business metrics.  The author instead advocates the use of “customer engagement” (Bingham, 2014).  Even when employing predictive analytics software, it is difficult to put a finger on customer loyalty he claims.   The author of this article, yours truly, believes we are experiencing a good old-fashioned semantic argument here.  Even the simplest RFM Analytics can easily determine loyalty, because a customer’s loyalty is based on historical behavior that is easily discernable.  Duh.  Mr. Bingham instead pontificates that “advocacy and involvement” are better predictors of customer loyalty and the resulting impact on revenue.  I have espoused the position to anyone who will listen that “you can’t eat engagement”, for many years.  Engagement doesn’t necessarily equate to sales, nor does advocacy.  Many people advocate the use of Leica cameras and are very engaged with the company via social media, forums, blogs etcetera, but have never made an actual purchase.  Engagement is no substitute for loyalty, and loyalty is most certainly an insight that can be determined through judicious use of Predictive Analytics.  Moreover, once the loyal customer is identified, his future behavior will be easier to predict and therefore he will be easier to communicate with.

      Michael Berry, Analytics Director for TripAdvisor.com takes the stance that “big data” is nothing more than hype (to which I agree), and that predictive analytics doesn’t require as much data as most people think.  He feels Predictive Analytics doesn’t require a marketer to figure out how to analyze all the data, but rather how to figure out how much data is needed to see something worth noting (again, agreeing here).  Berry states that testing the predictive model requires starting with what the marketer feels is just enough data, and then adding data incrementally so as to understand when enough is enough (Laskowski, n.d.).  I am frothing at the mouth in agreement.  Predictive Analytics is a tool that allows us to uncover hidden insights and trends, but doesn’t necessarily require every data point we own in order to provide useful information.  To expand on that notion, I would say that the vast majority of companies are too wrapped up in the pursuit of “big data” because it is the current fashion of the industry, but in reality they “need” far fewer data points to actually answer their pressing business questions.  But I digress.

     Of course he is.  In fact, the marketer may be more relevant than ever.  Marketing is a business conducted by people, toward people.  Computers and software, though incredibly helpful, can only process and understand the information they are given.  It takes a marketer to detect nuance, and to understand how a customer may respond to certain turns of phrase within copy.  Copy and copy testing is an integral part of the direct marketing process and one that relies heavily on the ability to communicate effectively.  That requires a human touch.  Moreover, a savvy, intuitive marketer is required in an organization because despite the massive analytical power companies have in the form of PA software, it takes a degree of experience and human understanding to know what questions to ask, how to interpret their answers and what potential information can be found through analyzing massive amounts of data.  Almost anyone can create a predictive model through inputting individual predictors, but only a marketer will know how to apply the results.  A marketer will know how to use the right information, at the right time, to solve the right problem.  Long live the marketer!

     In 2016, consumers are literally bombarded with marketing and marketing messages and as such they are typically only receptive to those that resonate with them on a deeper-than-average level.  Meanwhile, companies succumbing to the Siren call of Big Data, are collecting every shred of information about each customer they possibly can, resulting in gargantuan data sets

that would be impossible for a human to analyze with any accuracy.  The astute marketer needs to find a way to sift through and analyze the tremendous amount of data he has at his disposal in order to uncover the strategic advantages that will allow him to create that perfectly targeted, resinous communication good customers (and potential customers) will respond to.  Of course this is where Predictive Analytics comes in to play.   Software packages from SAS, SPSS, SAP and others can provide the marketer with more accurate information for making key business decisions and predicting the types of offers and calls to action that will yield maximum response with minimum marketing spend.  If asked, I would recommend that any company looking to identify trends, better understand their customers, improve business performance, drive strategic decision making and more accurately predict future customer behavior, should look into any of the top four PA software providers to see which one is best for them based on cost, scalability, and usability.   You still need your smart marketing people, but a PA package could make their lives a whole lot easier, and make their predictions a whole lot more accurate.

---Steve

Bingham, C. (2014, March 28). Customer loyalty is dead. Long live engagement. Forbes. Retreived from http://www.forbes.com/sites/oracle/2014/03/28/customer-loyalty-is-dead-long-live-engagement/

Halper, F.  (2014, March 4).  Four more advances in predictive analytics. TDWI.org Retrieved from http://tdwi.org/articles/2014/03/04/4-predictive-analytics-advances.aspx

IBM. (2010). Real world predictive analytics: Putting analysis into action for visible results. IBM Corporation. [White paper]. Retrieved on April 2, 2015 from http//www-01.ibm.com

Laskowski, N. (n.d.). Big data and predictive analytics: When is enough data enough? TechTarget.[Web log post]. Retrieved on April 3, 2015 from http://searchcio.techtarget.com/opinion/Big-data-and-predictive-analytics-When-is-enough-data-enough

Mariani, G., & Thompson, W. (2014). Big Data, Analytics and Hadoop. Retrieved

November 10, 2014, from

http://www.sas.com/content/dam/SAS/en_us/doc/conclusionpaper1/big-data-analyticshadoop-

107049.pdf

Par-Rudd, O. (2012).  Drive your business with predictive analytics. SAS. [White paper]. Retrieved on April 2, 2015 from http://www.sas.com/content/dam/SAS/en_ us/doc/whitepaper2/drive-your-business-with-predictive-analytics-105620.pdf

Predictive Analytics Today. (n.d.).  Top 27 predictive analytics software. Predictive Analytics Today. [Web log post]. Retrieved on April 5, 2015 from http://www.predictiveanalyticstoday.com/top-15-predictive-analytics-software/

Roebuck, K. (2011). Predictive analytics: high impact emerging technology –What you need to know. [1^st. ed.]. Berlin, Germany: Tebbo Publishing.

SAP.  (2015).  About SAP SE: Global company information. SAP. Retrieved from http://go.sap.com/about.html

SAS. (2015). Company history. SAS. Retrieved from http://www.sas.com/en_us/company-information.html#history

SPSS. (n.d.).  Enabling the predictive enterprise with predictive analytics. SPSS.com Retrieved on April 3, 2015 from http://www.spss.com.hk/predictive_analytics/work.htm (NEED)

Woo, S.  (2012, February 27). Online-retail spending at $200 billion and growing. The Wall Street Journal.  [Web log post]. Retrieved from http://blogs.wsj.com/digits/2012/02/27/online-retail-spending-at-200-billion-annually-and-growing/