LESSON - How Organizations Benefit from Predictive Modeling
By Rado Kotorov, Technical Director of Strategic Product Management,
Recently, there has been a surge in the use
of predictive modeling and scoring applications.
Contributing to this trend is the fact
that today’s transactional systems capture
massive volumes of complex data, with
certain elements having as many as 100
different attributes. However, only 5 to 20 of
these variables typically have an impact on
certain occurrences or conditions.
For example, income, education, and occupation
may affect a customer’s likelihood to buy
certain goods, but geography, gender, and
family status may not. The need to pinpoint
the most important variables, and what role
they play in certain events, is prompting companies
to adopt predictive modeling.
The ability to forecast the future, based on
the past, can help companies be proactive
instead of reactive to increase profits
and reduce costs. Consider the case of
inventory management. Organizations that
accurately predict future demand can maximize
revenues by having enough product
on hand to satisfy all customers. At the
same time, by precisely anticipating sales
levels, they can save money and eliminate
waste by avoiding stock surpluses.
Consider a credit card or loan company, which
can mitigate risk and financial loss by using
applicants’ histories to determine how likely
they are to default on future payments.
What You Need in a Predictive
Until recently, predictive modeling tools were
standalone packages that delivered robust
statistical analysis and data mining functionality
but limited BI capabilities. These solutions
were complicated to use, making them suitable only for researchers, statisticians, and
The true value of predictive modeling can only
be achieved when it’s leveraged by operational
users, who apply those same techniques to
the information contained in databases and
applications and use the outcomes to make
immediate operational decisions. To make this
possible, predictive modeling must be part
of a broader solution that seamlessly unifies
those features with data integration, reporting,
and Web access. Only then will operational
users be able to rapidly generate intuitive,
easy-to-understand, forward-looking results
through a user-friendly front end from information
contained in any enterprise system.
This is particularly important when you consider
that data preparation and manipulation
accounts for approximately 60 to 90 percent
of the cost of predictive modeling, and that,
due to a lack of the right skills, most statistical
research is never transformed into an application
(i.e., it remains a research paper). By
bringing these capabilities together, advanced
statistical analysis and forecasting can be conducted
in a far more cost-efficient manner and
be used by many more people, even if they
have no advanced statistical training.
Combating Cultural Resistance
Many companies experience push-back when
introducing predictive modeling. Management
is often the first group to raise objections, fearing
that the results will be complicated and
difficult to interpret, and therefore be unusable.
Those creating the models must ensure that
they are put in the proper form and business
context, which will make them intuitive and
Operational users are another group that
is often slow to adopt predictive modeling.
They worry the result will be counterintuitive,
uncovering mistakes in past decisions and
proving their gut instinct wrong. It’s important
to explain that predictive modeling will not be
used to hold them accountable, or to doublecheck
their choices, but to offer valid guidance
when making those decisions.
By allowing everyone to more intelligently
plan for the future, a predictive modeling
environment—one that is embedded within
a business intelligence infrastructure and
widely adopted by operational workers—can
empower companies to operate more efficiently
and profitably than ever before.
This article originally appeared in the issue of .