Social Media Analytics: Predictive Analytics

predictive analyticsPredictive analytics have been around a long time in finance and economics.  Slowly, these analytic tools are finding their way into the marketing and social media arenas.

What are predictive analytics?

Predictive analytics use data regarding past behaviors to predict how individuals will behave in the future.

For instance, your credit score is a predictive model including your repayment history and other information to predict whether you’re a good credit risk or not.

Predictive models commonly include a number of variables, such as # of late payments, and weighting factors that reflect the importance of that variable in predicting future behavior.  These are commonly regression-type models.

Modern predictive analytics use similar data and build similar models to predict how groups of people will behave, in general, or classify individuals into such groups.  For instance, we might build a model that predicts how much of a product we’ll sell if we lower (or raise) the price.  While we won’t be able to predict WHO will buy at the new price, we really don’t care.  We only need to know if we’ll sell more at the new price.  Thus, predictive analytics help us determine which marketing strategies will produce the best ROI (Return on Investment).

How businesses use predictive analytics

Businesses use predictive analytics in a number of ways, such as the one discussed above.  In addition, a number of tools, such as CRM (Customer Relationship Management) use predictive analytics to determine marketing strategies. Another type of predictive analytic is CLV (Customer Lifetime Value) which uses purchase information to classify customers into groups and determine the level of profit reflected by each group, which is used to build marketing strategies to each group.

Descriptive models and predictive analytics

Descriptive models are often overlooked as tools for generating predictive analytics because they suggest strategies that will generate better results without being able to quantify how much better the results will be.

An example is the TRA (Theory of Reasoned Action).  This model states that buying behavior is impacted by a consumers attitudes and beliefs about the products, as well as the norms related to that purchase.  This theory, of course, underpins how social media works. Social media helps build attitudes toward products based on the most credible sources — our friends — and establishes norms of behavior when we see all our friends buying the product.

So, why aren’t these descriptive models used more frequently in businesses.  In part, that’s due to poor exchange between businesses and academics who seem to speak different languages.

Predictive analytics and social media

Marketing in general, and social media marketing in particular, are not heavily influenced by predictive analytics.  Although, that’s changing as supercomputers allow organizations to use massive data captured during transactions to build predictive models of what consumers buy and factors that impact their purchases.

Still, relatively few companies use predictive analytics to drive marketing strategy.  Sometimes, when I pitch to prospective clients, I’m shocked at how few demand any true analytics from their agencies and almost none even understand the concept of predictive analytics.  If the agency provides any analytics, it’s commonly simple ones such as # of Fans, # or RT, or other somewhat meaningless data.

Agencies and in-house marketing employees often develop simple correlations as a way to build social media marketing strategy.  For instance, they might notice that certain types of content drive more engagement or that posting at certain times generates more engagement, so they do more of this.  But, this lacks to depth of understanding necessary to build predictive analytics.

I have several proprietary predictive analytics tools I use to help clients optimize their ROI.  For instance, I have a complex algorithm (model) to help businesses generate leads for the sales force from their email marketing programs.  In other cases, I build predictive models from scratch or use descriptive models to generate predictive analytics, such as my hierarchy of effects in social media.

Hausman and Associates

Hausman and Associates publishes Hausman Marketing Letter and the monthly email newsletter of the same name.  We also provide cost-effective marketing and social media through our virtual agency concept.  We welcome new clients and would happily provide a proposal to show you how we can make your marketing SIZZLE with predictive analytics.

You might also like:

Marketing Dashboards Track Marketing Strategy Success

marketing information system graphicsAccountability.

As economic times have strained purse strings in organizations, you hear the word accountability bandied around more and more.  Before adding anything to a marketing strategy plan, its ability to produce tangible, quantifiable results is required.  This can be a very good thing since there is never a good time to waste money on projects that won’t pay off.

Dashboards as a Tool to Monitor Success of Marketing Strategy

  • Dashboards form a graphic representation of data in a format easily interpretable by managers who need to use the data, but may not understand its importance or interpretation if displayed in a tabular format.
  • Information contained in the dashboard are related to achieving one or more marketing objective.  Examples of information that might be displayed are sales, margins, customer satisfaction, market share, Brand development index (BDI), stock-outs, etc.  These may be displayed as bar or pie charts and can be broken up by product line, SBU, etc.
  • Marketing strategy is evaluated by viewing results displayed on the dashboard.  For instance, a decline in sales over time might indicate the need to re-evaluate your current marketing strategy by increasing advertising, targeting messages more closely with the target audience, or reducing costs.  Meanwhile an increase in costs over time might suggest a different marketing strategy building on sourcing issues or re-design of the product with less expensive materials.  Multiple graphics on the dashboard help managers determine which change in marketing strategy fits with ALL the data supplied.
  • Information forming a marketing dashboard generally comes from internal records spread across multiple databases and, potentially, multiple functional areas of the firm.  For instance, sales might come from the marketing or sales department, while cost information may reside with accounting.  Some information will likely come from scanning outside resources, as well — doing market research.  For instance, market share relies on knowing the estimated size of the market, while BDI requires information on the size of a market segment.
  • Information on a marketing dashboard may be transformed using a variety of analysis techniques prior to graphing it.  For instance, regression, Logit, or other statistical analysis may be performed, then graphically depicted.

Characteristics of Good Dashboards

  • Remember that all information has a cost, so data collected and displayed on the dashboard should have some impact on decision-making.
  • Dashboards should be customizable, such that different managers have information most salient to their decisions.  Overwhelming managers with information they can’t use tends to encourage them to overlook more valuable information.  Managers at higher levels of decision-making should have overviews that contain summary information with the option to dive deeper into any area without re-configuring the entire dashboard.
  • Dashboards should not contain static information, but dynamic information such that data are plotted against time to show trends.

Software to Create Marketing Dashboards

A variety of tools are available for creating dashboards.  Among them are:

Cognos, from IBM


Pentaho (which is open source)

Other Resources on Building Marketing Dashboards


Demand Metric

Information Management

You might also like:

Marketing Analytics with SPSS

I spent the morning at a workshop hosted by SPSS (an IBM company) — thanks a bunch for the invitation.  Marketing analytics sure have changed in the last 20 years since I started using them — sure there’s still regression and logit and discriminant, but there are lots of new tools, too.

<p><a href=
Image: graur razvan ionut /

Today, I’ll give you an overview of the cool new modules and features of SPSS and, over the next couple of days, I’ll provide more details for each one.

Through development or acquisition, SPSS has a full line of tools for predictive analysis from data planning and data collection through analysis, reporting, and deployment of solutions.  The diversity of tools and ability to buy modules individually allows customizable solutions for a variety of business and non-profit problems.

A valuable feature of all the SPSS software is that the various modules fit together seamlessly and share a similar interface, making it easy to move between modules and reducing the learning curve.  The products also interface with other software products such as Excel.

Today’s session on predictive analytics dealt primarily with data collection, text analysis, new features of SPSS version 19, data mining, and Cognos Business Intelligence software.

Data Collection

This program, actually a group of related products, simplifies the process of designing visually appealing surveys and deploy them using wizards.  Surveys can be deployed through various media, including online and mail, and can be translated into various languages.

Text Analysis

Competing with more established products like nVivo and Hyperresearch, SPSS has developed a text analysis program that uses native language to search a variety of textual materials to find patterns in word use.  The categories can be collapsed into thematic patterns.  These categories can then be imported into SPSS Statistics and are coded as dummy variables.

SPSS New Features

SPSS is getting ready to release version 19, so they presented new features of the software.  Primarily, improvements have been made to make analysis run faster.  Another improvement is the ability to add text block to graphs removing the necessity of exporting graphs to a word processing program to add interpretations or other descriptions to the graphs.

Other changes have been made to make sharing data and output easier and allow for some modification of analysis by other users with only a browser.

Data Mining

Modeling software allows users to analyze vast amounts of data searching for unexpected patterns hidden within the data.  At lunch, someone asked how this differs from the basic SPSS data analysis and the answer is that it makes no assumptions regarding the relationships between variables while basic analysis requires at least a superficial understanding of the relationship, for instance, whether its linear.

Modeling software can handle text data in the way discussed above.  It can also export data to a server when more power is needed to analyze large data sets.


IBM recently acquired Cognos, which provides business intelligence (BI) to track a firm’s Key Performance Indicators (KPIs).  Cognos software provides highly customizable marketing dashboards to increase understanding of the data and displaying it graphically.

Additional Tools

In addition, SPSS offers new modules to handle direct marketing, which allows customized offers to customers based on past behavior, as well as calculates Customer Lifetime Value (CLV), which is important for grouping customers and determining how to market to them.  There are new tools to handle missing values and other data preparation tools to remove outliers, etc.  SPSS has also acquired AMOS, one of the top software packages for Structural Equation Modeling (what used to be called Causal Modeling) used for simultaneously evaluating constructs and relationships between constructs.

You might also like: