Empowering Business Analysts with Predictive Analytics

October 28, 2005

Empowering Business Analysts with Predictive Analytics

Empowering Business Analysts with Predictive Analytics

by Dr Fern Halper and Marcia Kaufman

Introduction

As part of ongoing research into trends in business analytics software, we have recently spoken with a number of companies who are working to empower the business analyst with predictive analytics tools – tools that were traditionally reserved for the statistician or computer scientist. Enabling subject matter experts who are not trained in predictive analytical modeling, to use such tools delivers more analytical power to more people. This could translate into significant top and bottom line revenue impact – but can this approach work?

The term predictive analytics emerged on the scene a few years ago and came out of the broader field of data mining and advanced statistical analysis. Predictive analytics tools help to identify patterns and establish relationships in large collections of data. For example, they are used by financial institutions to evaluate credit risk and prevent fraud and by pharmaceutical companies in drug discovery. A rough estimate is that half of the people who are using predictive analytics software use it on customer-related data to segment their customers and understand patterns of retention, churn and other behavior.

Predictive Analysis – Yesterday vs. Today.

Traditionally, predictive analysis was the realm of the statistician or the computer scientist who was well versed in artificial intelligence techniques such as machine learning or neural networks. Typically, a business analyst would work with a group of these highly skilled people to put together models that could be used by the business. These models would often be written using a scripting language. However, in many circumstances, this approach led to time delays and inefficiencies. Some of the most common problems were:

  • The number of statistical experts at a company is limited and often they could not keep pace with requests – leading to long wait times.
  • The business analyst requires multiple iterations to fine-tune a model based on business or market fluctuations. This adds additional time to a project.
  • The business analyst or decision maker may be closest to the business significance of the data, but does not have the ability to work closely with the raw data.
  • The statistical team may not focus in on the most important business issues because they may be distracted by the technical aspects of the problem.

Today, many software vendors are adapting their predictive analysis offerings to address the needs of the business user. The vendors recognize that enterprises are working with ever increasing amounts of both structured and unstructured data which adds to the complexity and the time constraints of completing a successful model. Vendors have also had requests from business analysts for years to make their software easier to use. This includes easy to use interfaces as well as pre-populating the parameters of the algorithms with reasonable default values that will return sensible results.

Vendors like SPSS and Insightful are stepping up to the plate to make this happen- albeit with somewhat different philosophies. The approaches of the two companies are summarized below.

SPSS

SPSS is one of the grandparents of commercial predictive modeling, with decades of experience in the field. Clementine, from SPSS, is an enterprise workbench that enables users to quickly develop predictive models using their own business expertise and deploy them into business operations to improve decision making. Clementine has been designed to appeal to both the statistician and the business analyst. The visual and interactive interface can be used with or without knowledge of a scripting language. The interface utilizes a flow approach, by which the user selects and connects icons from a palette. In order to improve the usability for business analysts, Clementine is designed to draw attention away from the technology to the data and patterns in the data.

Aside from the predictive modeling capabilities, the workbench includes facilities to get the data into shape for analysis. This includes data access, manipulation, cleansing, visualization, exploration, reporting and modeling as well as the ability to deploy the model. SPSS provides a way to get the results of the model out into the business ? with the scheduling of model runs. SPSS also believes that while tools such as Clementine may make it easier to do the analysis, domain knowledge is also required because the results need to be translated into the business domain. Someone needs to understand the relevant factors.

SPSS has recently added the ability to deal with unstructured (text) information in Clementine. It also provides both horizontal and vertical pre-packaged applications. For example, SPSS offers solutions such as PredictiveMarketing, and PredictiveClaims And, if the pre-packaged model is not appropriate, and the application is done frequently, Clementine provides what it terms ?Clementine Application Templates?. These are a set of ?stream? connected files showing how to do a particular application using Clementine plus sample data. It uses an icon-based interface. SPSS has also recently added ?Predictive Enterprise Services? which provides a central repository for analysts and business users to share analyses.

Insightful Corporation

Insightful Miner grew out of the S-Plus statistical package from Bell Laboratories. Historically, Insightful Miner was used primarily for research and prototyping, but it is now moving into production environments as well. Some of Insightful?s tools are similar to those provided by SPSS. Insightful Miner offers an easy-to-use graphical interface which the S-Plus scripting language sits beneath. Insightful also provides a set of tools to get the data into shape and to do text mining. Its philosophy for empowering business analysts is somewhat different from SPSS.

Insightful?s view is that those trained in predictive analytics/statistics should be encouraged to create the best possible models, which can then be used by business analysts (and others). The approach is to design a product that enables the modelers to create an interface for the subject matter expert who needs to use it. So, for example, a modeler might create a model for a marketing campaign and then a marketer would use this. Then, the modeler could create an application for that marketing group to use on the next campaign. This cuts down on re-runs and ad-hoc analysis, making everyone more productive.

Insightful believes that a company needs to decide how much of a competitive edge it wants- and it is reasonably certain that most companies will want the best model possible. So the idea is to develop these predictive models are then integrated them into a company?s business process. The approach has worked well for them in financial application areas and in drug discovery ? it’s a leader in both fields. For example, a biomedical company can integrate Insightful models with other tools that researchers use to identify biomarkers that will help with new therapies. These applications can be deployed to researchers as well as business decision makers. Another example; a marketing application can deliver the best model about what happened in past campaigns and tie that to a new campaign to enable better targeting of customers.

The Bottom Line

Hurwitz & Associates believes that both approaches described above have merit and that they can empower business analysts with predictive modeling capabilities. The reality is that predictive modeling isn?t easy. However, utilizing a visual interface with parameters preset at reasonable levels will produce reasonable output. The upside is that business analysts can produce sensible results and eliminate some of the traditional bottlenecks in the modeling process. The downside is that while the results may be sensible, they may not be optimal. For companies with a compelling need to create optimal models, for example in drug discovery or portfolio analysis, it makes sense for highly skilled people to either collaborate with subject matter experts or simply develop the models themselves. The approach a company takes will depend on what they are trying to accomplish.

The good news is that there are practical options out there for companies to explore. Hurwitz & Associates encourages companies to look at these options because the value proposition is quite compelling.

 

 

 

Newsletters 2005
About Dr Fern Halper and Marcia Kaufman, Partners

Leave a Reply

Your email address will not be published. Required fields are marked *