Behavioral Analytics

Behavioral Analytics for the Enterprise
Why, Where, What and How
Measure Everything
Web Data Streams
Web Analytics
Clustering Tribes
Engagement Marketing
Tell Your Friends
The Enterprise
Mob Profiles
Biofeedback Marketing
Consulting Services

Everyday consumers are bombarded with irrelevant marketing messages on the Web, email, wireless and other media.  Smart enterprises are starting to realize that to communicate over this fray of irrelevancy they must come to understand, leverage and model each of their customers’ behavior and preferences in order to provide personalized communications.  To accomplish and succeed at this objective, a behavioral analytics strategy is required by enterprises.   
 
One strategic move is to set up a series of behavioral analytical filters set-up by enterprises across multiple consumer channels – these filters represent conditions of events in the form of business rules – these conditional rules can be used to react and fire targeted offers of products, services or content.  These conditional business rules take the form of signatures representing profitable customers and/or opportunities for enterprises for growth marketing and targeted offers to consumers. The behavioral analytics filters can be strategically placed across customer touch points in the form of dynamic business rules.  These filters can easily and inexpensively be created using behavioral analytic tools incorporating machine learning algorithms.  These tools can be used to create dozens of decision tree which represent a series of predictive model.   
 
As with Internet mechanisms, such as cookies, forms, beacons, and JavaScript – decision trees can be the building blocks to a strategic behavioral strategy and framework.  Inductive decision tree software can be used to map from customer observations for targeting what kind of products or services to offer specific groups or classes of customers.   
 
More descriptive names for these decision tree models are classification trees – will buy vs. will not buy – that is they predict a discrete outcome.  This is an example of a business rule extracted from a classification decision tree for deployment by an enterprise:
   
IF         second visit to site
AND      IP Address Sector 5
AND      prior purchase price range $46.99 – $76.98
THEN    Will Buy 
   
Another type of decision trees are know as regression trees – how much will they buy – that is a continuous outcome. This is an example of a business rule extracted from a regression decision tree:
 
IF         second visit to site
AND      IP Address Sector 5
AND      prior purchase price range $46.99 – $76.98
THEN    Will Buy $98.03 – $99.87

   

A valuable source of decision tree papers, tools and consultants can be found at KDnuggets.com (Knowledge Discovery) a premier site of analytical information and software.  Decision trees are created using an assortment of algorithms and the cost can vary from a few thousand dollars to those that are free – here is a list from KDnuggets:  
 
AC2: a graphical tool for building decision trees.
Alice: a decision-tree-product for business users.
ANGOSS: decision trees for sales and fraud detection.
C5.0: constructs classifiers in the form of decision trees and rule sets. 
CART: multiple winner of decision tree (modeling) competition.  
DTREG: generates classification and regression decision trees.
Fair Isaac: tree-building software.
Neusciences aXi: discrete and continuous rules from trees.
PolyAnalyst: includes an information Gain decision tree.
Shih Tree Builder: regression and probability trees.
SPSS AnswerTree: CHAID and other decision tree algorithms. 
XpertRule Miner: graphical decision trees with embed ActiveX components.  

   

FREE:

     

C4.5: the "classic" decision-tree tool.
GAtree: genetic induction and visualization of decision trees.
IND: provides Gini and C4.5 style decision trees.  
Mangrove: visualization decision tree.
OC1: decision tree for continuous feature values.
ODBCMINE: analyzes ODBC databases using C4.5 decision tree.
PC4.5: a parallel version of C4.5 built with Persistent Linda (PLinda) system.
SMILES: decision tree with non-greedy search extraction.  
 
There is also software as service (SaaS) from Zementis.com aimed at providing enterprises with software at low-cost by eliminating commercial licenses costs. Zementis has a decision engine ADAPA(r) to accommodate enterprises’ needs for a flexible, secure, and incremental modeling solution.  Zementis offers an on-demand predictive analytics decision engine hosted on the Amazon Elastic Compute Cloud (EC2), it is based on Service Oriented Architecture (SOA) and open standards for model exchange.  Zementis is a cost-effective and a pay-as-you-go behavioral analytics service which provides a highly scalable framework to deploy, integrate, and execute complex decision models.  There are also software products that generate predictive rules directly from data which could be used as behavioral analytics filters.  Here is another list from KDnuggets: 
 
Compumine: rule-based predictive modeling software.
Datamite: enables rules to be discovered in relational databases.
DMT Nuggets: analytics based on Sift Agent(TM) technology.
PolyAnalyst: supports decision tree and fuzzy logic rules.
WizWhy: automatically finds all the IF/THEN rules from data.
XpertRule Miner: provides association rule discovery from ODBC data sources.  

     

FREE:

    

CBA: builds classifiers using a subset of association rules.
KINOsuite-PR: extracts rules from trained neural networks.
PNC2 Rule Induction System: induces rules using a cluster algorithm.  
  
Both decision tree tools and those listed here are capable of extracting IF/THEN rules which can serve as filters to issue alerts and targeted communications to customers by enterprises.  These filters or business rules describe the operations, definitions and constraints – as well as the opportunities and conditions which can be applied to enhance growth and revenue for enterprises.  The behaviors of consumers provide the framework for applying these analytical tools to derive rules.   
  
There are also analytical tools which incorporate multiple algorithms with highly sophisticated interfaces they include the following software suites: 
  
ADAPA® from Zementis: a framework for deployment, integration, and execution of various predictive algorithms, including neural networks, support vector machines, regression models, and decision trees.
Clementine: from SPSS, visual rapid modeling environment for behavioral analytics.  
KINOsuite PR: extracts rules from trained neural networks.
Knowledge Studio: featuring multiple models in a visual, easy-to-use interface.
MarketMiner: automatically selects the algorithm: statistical networks, logistic and linear regression, K-nearest neighbors, and decision trees (C4.5).
Mathematica: multi-method system for computational models from data.
Oracle 9i Data Miner: embeds into Oracle9i database, for making classifications, predictions, and associations.
Polyanalyst: multiple classification algorithms: Decision Trees, Fuzzy Logic, and Memory Based reasoning.
Predictive Dynamix Data Mining Suite: integrates neural network, clustering, and fuzzy models.
PredictionWorks: includes decision tree, logistic and linear regression, etc. 
Previa Classpad: neural networks, decision trees, and Bayesian networks.
prudsys DISCOVERER: decision trees and sparse grid methods for classification.
Rank from VADIS: multiple behavioral analytics algorithms software suite.  
STATISTICA Data Miner: multiple modeling algorithms.
Tiberius: neural networks, logistic regression, 3D visualization, etc.
  
Most of these software suites offer a comprehensive selection of algorithms for automated analysis of text and structured data. Numerous data analysis problems in various application fields are readily solved by these types of behavioral analytical tools, enabling users to perform numerous knowledge discovery operations, such as categorization, clustering, prediction, link analysis, keyword and entity extraction, pattern discovery and anomaly detection.
  
Imbedding these filters at websites, call sites and other operational systems can be used to build and manage customer relationships.  Recognize that developing, testing and deploying these filters is a learning process conducted in an iterative fashion.  The end result will be twofold: knowledge discovery and improved growth for enterprises.  The insight gained will influence strategic direction as well as improved relevance in how enterprises communicate with each of their customers. 
  
The need for generating and evolving a set of filters based on relevant customer behavior patterns in enterprise data is fundamental to enabling personalized communications and targeted customer services across all channels.  Effective behavioral analytical solutions require a set of inductive business rules that facilitate the automation of processes, the framework for accomplishing and leveraging these filters should be flexible and ongoing as conditions change.   
  
The collection of consumer demographics can strategically be accomplished several ways: first, it can be directly solicited by forms – importantly the consumer should be informed of the value of providing their personal information: relevant information streamed their way. Demographics are important in that they provide vital data points on a consumer’s life cycle, some will be looking for baby cartridges while others retirement properties in Mexico. Demographics can also be gathered indirectly for example; a store locator form soliciting a ZIP code can yield the following information:  
  
Find Our Store – Your ZIP Code: 79902 
    
ZIP Match: Southwestern Families Cluster 
Core Market Driver:  Two kidsConsumer Profile: Blue-collar/service occupation, median age 28.6 years/income $27,327.
Product Offers: Children's products, car replacement tools and parts Marketing Channels: H.E. Butt, Albertson's, Vons, Hispanic radio and the Web. 
   
From a ZIP code we now know the level of interest, price, placement, products, and language this consumer cluster is coming from. The offers and ads presented to them should take this into consideration.   A more specific profile can be generated from a consumer address, and household demographics from such firms as Claritas, Equifax or Trans Union.  Aside from enhancing the precision of offers based on lifestyle consumer clusters, there is also the leveraging of the Geolocation of their IP address.

Behavioral analytics may also involve the integration and analysis of online and offline information.  This includes a mixture of offline demographics and online clickstream data as well as wireless WAP type of information.  Two major traditional data aggregators Acxiom.com and Experian.com are offering streams of real-time demographics and lifestyle information for behavioral analytics to marketers and advertisers, as well as enterprises. 

These demographics data products are design to predict consumer response and life time value.  Marketers and advertisers can use these real-time demographics to enhance their cookie, registration and other operational databases in order to improve their online sales via segmented behavioral analytics based on a combination of online behavior and offline lifestyle clusters.  Demographics providers are merging their household clusters of millions of US households in the form of networked cookies to target online ads for behavioral analytics.  
   
These new demographic cookies offerings contain lifestyle information allowing an enterprise which subscribes to these demographic networks to target specific ads for specific lifestyle groups. So that for example, those in the household cluster of say “Cartoons & Carpools” would get shown online ads for minivans rather than sports cars. Or, those consumers coming from a ZIP code with a high concentration of apartments would be shown an ad for a window air conditioner rather than one with remote controls designed for central heating.  
  
Experian real-time marketing service validates a consumer’s name and address against their own information and returns relevant data for websites to use, including demographics and diagnostic information.  The Acxiom product is Relevance-X an online advertising network dedicated to deliver relevant offers based on the uses of their demographics to the right consumers.  Acxiom Digital is also broadening its Internet marketing solutions through the acquisition of Kefta, a Web site personalization company. 
 
Relevance-X offers consumers a choice about participating in this experience including information on how to delete and block cookies from being set to their computer. Relevance-X offers segmented results based on demographic, lifestyle factors and shopping patterns. The platform is the result of a blending of Acxiom's EchoTarget ad network (which Acxiom acquired) and PersonicX their targeting system based on 70 unique household segments. Both Experian and Acxiom clearly see the value of the convergence of their offline demographics and the various lifestyle clusters with that of clickstream shopping patterns. For the advertiser and enterprise they offer two options in leveraging the information they provide, the subscription model as well as the enterprise model.  

 

Enter supporting content here