|
1.
TRAINING SCHEDULE UPDATE
COURSE SERIES ON DATA MINING STRATEGY, METHODS AND APPLICATION
Learn how experts
mine data by attending The Modeling Agency's vendor-neutral,
application-oriented data mining courses. Participants
will enjoy a balanced and broad presentation of predictive
analytics without restriction through a particular tool or
product. Attendees will learn about data mining
capabilities, limitations, methods, tools, strategies,
techniques, applications, and costly pitfalls. Those in attendance
will leave with a comprehensive binder of notes,
illustrations and references to valuable resources.
Don't leave a
powerful competitive advantage untapped: harness the valuable information and
profits hidden in your data. Each offering is limited to just 18 seats. A current status of remaining space may be
viewed at TMA's main training page.
Submit an
unofficial registration and reserve your seat today while
your training request is processed.
Since The Modeling Agency is not a
tools vendor, participants enjoy a balanced, broad and
non-promotional perspective of predictive analytics at desirable
venues throughout the USA. |
CHOOSE THE TRAINING
THAT'S RIGHT FOR YOU
The Modeling Agency offers three data mining courses with
distinct objectives. The courses are designed to be
attended independently, or as a progressive series. While the
three levels are staged as a progression, they should not be viewed
simply as "introductory, intermediate and advanced." Refer to the table
below to ensure that your experience, situation and objectives align
properly with the intent, scope and depth of each offering:
|
Course |
Focus |
Scope |
Geared To |
|
Data Mining:
Level I |
Strategy |
An intensive overview
of strategy, best practices and case studies |
Project
leaders,
Stakeholders,
Functional Managers |
|
Data Mining:
Level II |
Methods |
A tactical drill-down
of the data mining process, methods, techniques and
resources |
Business
Analysts,
Functional Analysts,
IT Professionals |
|
Data Mining:
Level III |
Application |
A hands-on
application workshop as an extension to Data Mining: Level
II |
Practitioners,
Model-builders,
Decision Support
Developers |
|
|
2.
FEATURE ARTICLE
SIX KEYS TO RECOGNIZING AN
ROI WITH PREDICTIVE ANALYTICS
by
Christopher Checco and Bill Kastner
DHSoft
INTRODUCTION
According to a recent IDC study1,
the use of Predictive Analytics yields a median ROI (Return on Investment)
of 145%, which almost equates to double the ROI when non-predictive
analytics is used. Predictive Analytics is now available to the masses and
has become a distinct competitive advantage for those organizations savvy
enough to implement it. For those who have not yet taken the plunge, the
time has come to take the steps, as Predictive Analytics will become a
necessity to compete by the end of this decade. In fact, all ten processes
identified by the Gartner Group in the article “Top-10 Marketing Processes
for the 21st Century2” would be well served by Predictive
Analytics, either directly or indirectly. The recurring theme in these 10
marketing processes is building stronger relationships between organizations
and their customers. Predictive Analytics goes a long way in addressing
this challenge.
WHAT IS PREDICTIVE ANALYTICS AND HOW IS IT UTILIZED?
Simply stated, Predictive Analytics is the
implementation of statistical modeling to generate ranked lists based upon
propensity to exhibit a certain behavior. What does that mean to business?
Predictive
Analytics can be used in various commercial arenas in many ways. Following are
a few examples to help clarify how Predictive Analytics is used to help
organizations:
-
Customer Retention –
Predictive Analytics can help organizations identify which customers are
likely to churn (cancel service, stop using a product, etc.). Additionally,
it can identify likely causes for the attrition at the individual customer
level
-
Customer Acquisition –
Predictive Analytics can help identify which prospective customers should be
targeted. Furthermore, it can identify which specific offers are likely to
be effective, as well as estimate future customer value.
-
Cross-Selling and
Up-Selling Opportunities – Predictive Analytics can aid organizations in
identifying which products and services individual customers are likely to
buy. Furthermore, it can help identify future profitability of individual
customers who add specific products or services.
KEY BENEFITS OF PREDICTIVE MODELING
So why is Predictive Analytics so effective?
There are several reasons and here are the two most relevant:
1)
Self-Improving – As the organization learns about the key factors that
affect their business through the use of Predictive Analytics, they become more
in tune with their customers. This, in turn, allows them to gather more
accurate data to use in the predictive modeling process, making the results more
fine-tuned with each pass. Hence, marketers can easily recognize the positive
impacts, such as improvements in response rate or reductions in customer
attrition.
2)
Measurable – Predictive Analytics allows organizations to measure
effectively key metrics that feed into ROI analyses. This is mainly due to the
organized manner in which Predictive Analytics embeds itself into both business
and IT processes.
Unlike
traditional data segmentation, which relies heavily on demographic data,
Predictive Analytics focuses on individual customers by taking into account the
behavioral patterns of individuals. The differences between the two are
displayed in Figure 1.
This is accomplished
through the analysis of large volumes of behavioral data within the modeling
process (versus a limited amount of demographic data). The data is thoroughly
investigated until hidden patterns are revealed within the data, which
transforms raw “data” into useful and actionable business information.
Predictive Analytics is
often measured using a concept called “lift”, which represents the increases in
the response (or take) rate over and above the current hit rate. In Figure 2, a
marketer is striving to address a customer need through the use of a
product-based campaign. The marketer conducts a conventional analysis by
dividing the customer universe into demographic segment, and sends the offer to
50% of the customer universe. Over the coming weeks, the marketer calculates
that the response rate was 5% as represented on the left side of the Figure 2.
During a subsequent campaign for a similar product, the marketer employs
Predictive Analytics. The results of the Predictive Model show that the
marketer should target a different sub-segment of customers, which equates to
30% of the customer base (down from 50%). However, this sub-segment is not a
demographic segment of the customer base, but rather is characterized by a
complex combination of behavioral, environmental, and demographic data. In
tracking the offer, the marketer realizes a 14% response rate, which translates
into a lift of 2.8.
One can derive that
Predictive Analytics yields two major benefits to the marketer. A comparison of
the benefits is made in Figure 3.
1)
It reduces the number of customers an organization must target, which
reduces the total marketing expenditures (or allows funds to be shifted to other
marketing campaigns).
·
Using Conventional Analysis, the
marketer sent out 5 million offer letters at a cost of $1M.
·
Through the use of Predictive
Analytics, the marketer was able to send out 3 million offer letters at a cost
of $600K, netting a savings of $400K.
·
The savings equate to a 40%
expense reduction through the use of Predictive Analytics.
2)
It increases the number of respondents, allowing for a greater return on
the investment.
·
Using Conventional Analysis, the
marketer realized a 5% response rate yielding a total of $1.2M in additional
revenue.
·
Through the use of Predictive
Analytics, the marketer was able to increase the response rate to 14% generating
$3.36M in revenue, which equates to a $2.16M increase in net revenues.
·
This equates to a 180% increase
in revenues and the response rate through the use of Predictive Analytics.
One may ask, “With such
significant, quantifiable gains, why would anyone not take advantage of this
technology?” DHSoft’s research shows that most companies do not use Predictive
Analytics for the following reasons:
-
There is a perception
that the technology used for Predictive Analytics is cost prohibitive.
-
There is a lack of
understanding regarding the potential uses and depth of the technology.
-
Some feel that they
are already utilizing Predictive Analytics (via traditional reporting and
segmentation).
-
There is a
misconception that Predictive Analytics will reduce the potential value of
the marketing professional.
-
There is doubt that
the results are being accurately measured.
-
There is an
underlying fear of the unknown.
-
Traditional incentive
programs reward based upon gross sales, rather than rewarding based upon an
increased efficiency rating (such as Return on Investment).
CRITICAL FACTORS FOR SUCCESS
Predictive Analytics is an
extremely powerful tool; however, organizations must take a structured approach
to the planning and implementation of the infrastructure:
-
Include Business
Insight – Predictive Analytics, like any other tool, must be implemented
with a thorough understanding of the business that it is serving. This is
especially true when validating the results of the model, as many times data
attributes which are identified as key indicators are nothing more than
noise. Utilizing a subject matter expert (SME) will allow for a proper mix
of intuition and analytics when reviewing model results.
-
Have a Tactical Plan
to Utilize the Results – A common output of a statistical model is a ranked
list of customers that have a propensity towards some specific behavior
(churn, cross-sell, etc). These lists can be used to drive marketing
campaigns, however, in many cases, the users (e.g. – marketing and sales)
are unsure how to utilize the lists. In a recent conversation, one CEO
indicated that they were receiving ranked lists from a vendor; as the
conversation progressed, he confessed that they had no methodology for
utilizing the lists in any constructive manner. Organizations must have a
predefined process to take the results of the modeling process and use it in
an effective and efficient manner to support business objectives.
-
Review the Model
Results Regularly – Many organizations that use Predictive Analytics fail to
implement two simple and necessary steps:
-
Model Validation – This is the
process of ensuring that the behaviors one believes have been predicted are, in
fact, valid. This is done simply by testing the results of the model on a
sample set of data and reviewing the results. This quick and painless check can
save money and provide confidence that the process will provide value.
-
Model Degradation – The process of
Predictive Modeling requires that the results of a model be monitored
over time, to measure the effectiveness of the model as change occurs.
There are two common themes within the realm of model degradation that
should be considered. The first theme is one of a natural
degradation process, in which the target audience shrinks over time as
they respond to a particular offer. For example, in the
telecommunications industry there may be a case of a cross-sell campaign
targeting customers who do not have caller-id service on their home
phone. In the first month, the offer may go to 100% of the target
audience (for simplicity sake, target all customers that do not have
caller-id service). If a positive response is received from 15% of
this audience, the available universe shrinks to 85% of the original
target (assuming no new customers are added that do not have caller-id.
As more responses are received, the effectiveness of the model will
naturally degrade because of this shrinking target universe. The
second theme focuses on the changing influence of variables used within
the model. An example of this would be the influence of rising gas
prices on consumer behavior. A significant change in gas prices
may alter consumer behavior, potentially changing the way gas prices
influence the model. Hence, the effect of the gas price attribute
within the model needs to be refreshed to reflect the impact of the
change in price. One way to measure the usefulness of a model is
to test it periodically against known record sets and map the results
over time. In some cases, a simple model refresh will recalibrate
the model to reflect the ever-changing environment; in other cases, the
model will need to be retired.
-
Integration of Key
Metrics During Implementation – Predictive Analytics can become a powerful
marketing tool, especially when it is used with key business metrics. Some
examples include the use of metrics such as Customer Value, Lifetime Value,
or Credit Risk along with their ranked lists to make informed, fiscally
responsible decisions regarding offers. For instance, if a mobile phone
customer that generates $2 per month in profit is at risk of churn, it does
not make sense to offer that customer a $200 phone upgrade, even with a
2-year contract extension. However, if that given customer that generates
$30 per month in profit, that same offer would make fiscal sense.
-
 Take a Phased
Approach – It is important to remember that Predictive Analytics is a
complex, iterative process that requires periodic measuring and tuning.
Because of these factors, it is a wise choice to take small steps when
implementing this toolset into an organization. This allows for minor
changes to be implemented on an incremental basis with minimal impact, and
the results from each project can be rolled into subsequent operations. In
line with this thinking, organizations should start this process by picking
a small goal, and stay focused on that goal throughout the implementation
cycle.
-
Balance the Art and
the Science – There is an age-old conundrum between statisticians and
marketers; in general, statisticians tend to be quantitative and marketers
tend to be qualitative. Each group has its own strengths, but the sum of
the two parts is certainly greater than the individual pieces.
Organizations must strike the balance between the Art of Marketing and the
Science of Predictive Analytics to be truly successful.
CONCLUSION
Predictive Analytics has
been brought to the masses, and is no longer a tool just used by
mega-corporations and government entities. Predictive Analytics is a powerful
tool for many organizations, and can be used across industries to help drive
efficiencies into the business. If organizations use care in planning and
executing, success will be imminent and the financial returns that once seemed
so unreal will become a reality. As strong customer relationships become
increasingly important to companies and their customers, Predictive Analytics is
there to help, as it:
-
Allows companies to
gain an in-depth understanding of their customers needs, bringing the
customer-centricity of the “Mom and Pop shop” to companies of all sizes.
-
Provides the
knowledge and insight for organizations to build those ever-so-important
relationships with their customers by anticipating their needs.
-
Addresses customers’
needs at the individual level based upon the customer’s behavior (something
that simply cannot be accomplished through traditional segmentation).
-
Increases the
efficiency of the marketing ROI, allowing organizations to focus resources
on other critical areas.
ABOUT THE AUTHORS
Christopher Checco is the Vice President of Modeling and
Analytics at
DHSoft, Inc. He brings over ten years of experience as
a business and technical consultant. As a Technical Project
Manager, he has successfully implemented a number of large
software integration projects, in various industries.
 Prior to Prior to joining
DHSoft,
Mr. Checco worked as a Project Manager and Systems Analyst
for a variety of companies including The Carpe Diem Group,
Phillips Publishing, and MEI Software. In addition to being
a Certified Project Management Professional, Mr. Checco also
has an International Executive MBA from Georgetown
University in Washington, DC.
Bill Kastner is a statistician at
DHSoft,
Inc. He brings over fifteen years of experience in
statistical analysis, data management, and database user
training and multi-vendor computer system development. Mr.
Kastner has spent the last two years providing Statistical
Analysis to Nextel.
Prior to joining
DHSoft,
Mr. Kastner provided analytical services to many clients
including AT&T, DoubleClick, and Rockwell International. He
has a Master’s degree in Statistics from Colorado State
University in Fort Collins, Colorado.
1
Predictive Analytics and ROI: Lessons from
IDC's Financial Impact Study by Henry Morris. IDC September 2003.
2
Analyst View - Top-10 Marketing Processes for the 21st
Century by Gartner Research Analysts Claudio Marcus and Kimberly
Collins. CMO Magazine. July 2005.
Published
with permission from
DHSoft.
Copyright ©
2006
3.
ARTICLE
TEXT MINING FOR BUSINESS APPLICATIONS
by
Curt Hall
Senior Consultant
Cutter Consortium
INTRODUCTION
Over the past year, I've noticed increasing
attention being directed at the use of text
mining to enhance customer relationship
management (CRM), knowledge management (KM),
enterprise information portals (EIP), and other
applications by automating the analysis,
categorization, indexing, summarization, and
association of high volumes of text-based
information.
Although text-mining technology has been
available for years, its use has mainly been
relegated to intelligence agencies, news
providers, wire services, and other
organizations whose business largely revolves
around handling or processing textual
information. However, the corporate adoption of
the technology for more general business
applications has been limited.
One of the major issues with text-mining
technology has been the degree of expertise and
learning curve required for building
applications. In particular, creating the
directory structure or taxonomy necessary for
capturing unstructured and semi-structured data
from various customer touch points (customer
comments in e-mail, etc.) and integrating it so
that it can be analyzed has proved beyond the
capability of many rank-and-file businesses.
This has been aggravated by a lack of suitable
taxonomy and other classification tools.
Consequently, application development has
frequently required the use of highly
specialized consulting services, making
implementation costly.
Although many issues still remain with text
mining technology, it appears that organizations
are beginning to feel more comfortable using it
-- or are at least willing to investigate text
mining. Several developments are responsible.
For one, text-mining products have advanced over
the past few years with the development of new
visualization and user interface techniques.
General advances in XML have also helped to ease
integration and manipulation of textual
information. In addition, the technology is
becoming embedded in other software
applications. The result is that I am now seeing
text mining used for business uses, such as:
-
Research -- to discover previously
overlooked relationships contained in
volumes of biomedical literature, warranty
information, and other large collections of
text
-
CRM -- to automate analysis of large volumes
of customer e-mails and other business
correspondence
-
KM -- to automate the interpretation of
text-based business information to
categorize and organize it according to how
it relates to other corporate resources such
as documents and employees
-
EIPs -- to automate development of corporate
portal taxonomies (i.e., the directory
structure that end users use to browse and
search for information on the portal
I have also noticed the appearance of
conferences dedicated to the business
application of text mining, as opposed to just
technological developments. For example, the 2nd
Annual Text Analytics Summit, scheduled to take
place 22-23 June 2006, in Boston, Massachusetts,
USA, will feature sessions covering how
companies such as The Hartford Financial
Services Group, Hewlett-Packard (HP), Procter &
Gamble and others are utilizing text mining (see
http://www.textanalyticsnews.com
).
CONCLUSION
Text mining is
definitely receiving more attention by companies
for use in more mainstream business
applications. Just exactly how much is still
uncertain at this time. Over the years, our
surveys have consistently shown that less than
20% of end-user organizations use text mining;
however, we are now preparing a new survey
designed to measure current use of text mining
and I will detail our findings in upcoming
reports. Until then, I welcome your comments and
opinions on the practical use of text mining
techniques. Send your comments to chall@cutter.com
or call me at +1 510 848 7417.
ABOUT THE AUTHOR
This excerpt, by Cutter Consortium Consultant Curt Hall, originated
from Cutter's Business Intelligence Advisory Service. Through this
subscription-based service, you are assured of expert analyses of
the latest business intelligence strategies, products, and technologies.
For more information or to find out how you can become a client,
please visit
Cutter Consortium's web site,
or contact Dennis Crowley at + 1 781-641-5125 or e-mail
dcrowley@cutter.com.
Published
with permission from
Cutter Consortium. Copyright
© 2006.
|
