NUMERIC COMPONENTS AND DATA MINING PART III
by NAG for DSstar
Numeric Components and Data Mining Part III: Lessons from PeopleSoft
Q: What drives performance and profitability for your company? What are the
revenue sources, components of expenses, and opportunities available to
optimize shareholder value?
A: While developing their corporate strategies, these are the types of
that users of PeopleSoft 8 ask and find answers for using its powerful
analytic tools for enterprise performance management. PeopleSoft Inc in turn,
asked themselves how they could best bring these new analytic tools to their
users. PeopleSoft concluded that seamlessly weaving the world's most tested
and proven statistical algorithms into their applications would provide
compelling advantages to their company. In this article, Louis Olds,
PeopleSoft's vice president of Product Management explains how numeric
components are used in their highly acclaimed tool for automated knowledge
Q: What types of automated knowledge discovery does PeopleSoft 8 provide to
A: PeopleSoft Enterprise Performance Management is a comprehensive
suite of eBusiness analytic applications designed to help companies manage
their investments in employee, customer and supplier relationships.
For example, a company would be very well-served by knowing whether it sells
more or less via the Internet, or in stores, to particular market segments
such as yuppies or seniors, and so forth. In fact, most large companies have a
vast warehouse of sales data that they can plumb for answers to such
questions. PeopleSoft's applications allow users to conduct mission-critical
analyses such as these that can have deep impacts on profitability and
Q: How do the statistical and mathematical components in PeopleSoft's
Enterprise Performance Management enhance this capability?
A: PeopleSoft has embedded mathematical and statistical components from
C Library into PeopleSoft 8 Enterprise Performance Management products. This
gives PeopleSoft users the advanced statistical tools they need to effectively
analyze information. These components are seamlessly woven into our tools for
asset liability management, funds transfer pricing, risk weighted capital, and
Q: Are these tools then specifically for the financial services industry?
A: Many users are in fact in the financial services industry but there are
many other types of companies that can and do use PeopleSoft 8's analytical
tools with NAG components embedded into them. For example, companies with long
term contracts that are very much affected by customer behavior (such as
deregulated energy companies, wireless telephone companies, etc.) are also
likely to find these as powerful ways to manage market risks.
Typically users are vice presidents of management accounting, corporate
planning, or other executives working with company CFOs. The statistical
components are embedded deeply into the applications and users are only
required to know the business rules and assumptions that drive an overall
analysis. We have abstracted away as much of the technical complexity as
possible. This means that the software is maintained by business analysts, not
an IT Department, and is one of the ways in which these analytical tools make
a break with past practices. In the past, companies usually had sophisticated
analytics in a black box written in code by their IT department and there was
no way to update how calculations were done unless they went back to the IT
department in their company.
The users of the financial services applications where NAG code is embedded
are often from very large institutions that need to calculate risk adjusted
profitability in fine detail, often down to the individual customer among tens
of millions of customers. In the past, such companies probably tried to use
spread sheets and calculating risk got unwieldy in very short order. Now, for
example, a bank with 100 different business units worldwide can keep track of
how every loan it makes contributes to the risk in its overall portfolio. They
need statistical tools to make these analyses.
Q: Was the size of PeopleSoft users' datasets important to your decision to
use NAG statistical components?
A: Yes, it was one of the key decision points. Our financial services
clientele are calculating risk adjusted profitability for many millions of
customer transactions. This made high performance and scalability very
important factors in our consideration of which tools could be embedded in our
We also knew we needed sophisticated math routines that would, like
PeopleSoft's analytic applications, work with a number of relational database
environments, both for Unix and mainframes. PeopleSoft's eBusiness analytic
appications allow our customers to leverage the investment they've made in
their computing environement.
Q: Why didn't your development staff create these math routines internally?
A: We certainly could have developed these routines IF we wanted to take
time to do so. We have very sophisticated mathematicians on our staff but we
wanted to avoid the time to reinvent the wheel by writing the lower level math
ourselves. We estimate it would have taken more than two man-years to do this,
which is an awful lot of time and expense.
We looked at a few different math libraries and found NAG by far had the
better set of routines that we needed. Their C Library was proven to be more
accurate and with the highest performance. NAG has the credibility we required
from a tool supplier to move forward with confidence as we built our
In summary, using NAG components enabled us to get to market very quickly with
fairly sophisticated analytics. None of our customers know that NAG exists in
the product because it is abstracted away from those complexities and put in
terms of business analytics. Nonetheless, our users get the power and
flexibility of these NAG routines. We expect to reduce maintenance costs by
using such reliable and proven code. This will allow us to focus on new
functionality in future releases.
Q: What advice do you have for companies considering the use of numeric
components in their applications?
A: My advice to others seeking numeric and statistical components is to
for libraries with the breadth of algorithms you need and with
well-established accuracy. Also keep your eye on the efficiency of the
routines, ensuring that you find the most efficient algorithms that have been
developed for those functions.
For more information on NAG numeric components, contact Tony Nilles, Numerical
Algorithms Group, 1400 Opus Place, Suite 200, Downers Grove, IL 60515-5702,
630-971-2337, FAX 630-971-2706, firstname.lastname@example.org.