BY MARK LEON
Senior Technology Writer

SAN FRANCISCO - Mining corporate data for strategic advantage is still an esoteric art, but KPMG Peat Marwick hopes to win new accounts and leverage old ones with its new Center for Data Insight in Flagstaff, Ariz., where KPMG can use the resources of Northern Arizona University's Computer Sciences and Business departments.
     "The University really runs the center on a day-to-day basis," says Steve Cranford, KPMG global partner in charge of Data Warehousing. "NAU is focused on getting their name on the map, and our clients like the academic environment. And KPMG also likes the fact that a pool of cheap intellectual talent comes with the academic environment in the form of master's degree candidates. 
Data mining requires it. A mix of Artificial Intelligence, advanced, multivariate statistical analysis, neural networking and good old relational database management, data mining is not for the commodity
IT professional. "This [data mining] is the last bastion of data warehousing, says Cranford. "It's coming on the heels of OLAP [On-line Analytical Processing] , ROLAP [Relational OLAP] and MOLAP
(cont. below)

[Multidimensional OLAP] market.  "In setting up the center, KPMG sought to create a facility where a client could see a blueprint for building a "knowledge management system." What Cranford calls "knowledge management" refers to a practical set of sophisticated techniques that a corporation  can apply to a data warehouse to gain competitive advantage.
    Analysts disagree on what distinguishes data mining from more standard OLAP. "You don't have to differentiate data mining from any form of analytics, " says Alexis dePlanque, an analyst with the Meta group in Stamford, Conn. "But we use more stringent criteria, defining data mining as analysis that incorporates artificial intelligence and hypothesis generation."
     Not all of what KPMG does at the center would meet this test.   Credit card scoring, which Cranford cites as an example , does not qualify under dePlanque's definition. KPMG, however, is pushing the envelope of what enterprise integrators normally do in this area precisely because it is so new and poorly defined.
     In order to give clients an opportunity to test the waters, KPMG has created a number of engines or analytic models. These models, running against a data warehouse, will provide insight into profitability, customer risk and customer segmentation.  "We have used SAS Institute's software to build these analytic engines, " says Cranford.  "We source the data from a variety of systems to create the data warehouse and possibly a number of smaller data marts."
     One of the first clients to use the center after it opened in December was Mobil Oil Corp.  The first phase, loading the data, was easier in that case because it came from Acxiom, a Conway, Ark., company that Mobil uses to maintain good, clean customer data.  Before any analytics can be run against a data warehouse that data must be clean, free from most of the errors that  often creep into transactional data.  " The data we get isn't always in such good shape," notes KPMG's Cranford.  "we often have to do some of the cleansing ourselves."
     Cranford adds it's important for a client to come to the center with a well-defined set of problems.  Mobil's card business manager, Lenn Eason, says he hopes to benefit from the engagement in three areas: Developing better customer acquisition strategies, better retention strategies and more reliable credit risk prediction.
       Says Cranford, "In the past, this information usually cam from traditional market research which is self-reported, and hence not that
   

reliable."  Data mining, according t Cranford, will give Mobil actual patterns, and the analysis can be done at a very granular level.
    Analysts, however, caution against putting too much faith in data mining.  "Initially, corporations though you could do retail basket analysis with it[data mining] , "reports dePlanque.  "And while you can identify customer behavior, there isn't always an actionable solution.  A year ago we thought customer churn in telecommunications would be a big market.  But even if you find a customer who has switched carriers, what action do you take?  Do you send another flyer?"
     Data mining software vendors, according to dePlanque, have had to recognize that the market for generic data mining tools is negligible.  "You have to come up with packaged solutions which solve a specific problem.  For certain business problems it's valid."
     The Flagstaff CDI is partly a response to these kinds of questions about data mining.  By partnering with NAU, KPMG has created a place where clients can test the waters without making a huge investment.  "KPMG approached us to see if we would be interested," says Eason.  "This project is a good, low-cost way for us to get into data mining."
     Cranford says the center is also a good place to test the various data mining tools that are available.  "We try to be objective in our evaluation of tools.  Ease of use is increasingly important.  Corporations prefer that the tools no require a Ph.D. to operate."
     The top mining tools that are available at the center include software from Thing Machines, Inc. of Bedford, Mass., and Enterprise Miner from SAS Institute of Cary, N.C.  "IBM's Intelligent Miner is not yet a preferred partner, " Cranford adds.  "There are issues of capability we are not yet comfortable with."
     Given the interest in data warehousing, analysts were predicting a boom in data mining.   But dePlanque says the numbers haven't borne out that belief.  "The smaller data mining companies are seeing serious stagnation, "he concedes.
     Even so, large corporations are searching for ways to get more competitive advantage out of their data, and data mining is still the newest way to do that.  However, clients should be wary of large, loosely defined projects, warns dePlanque.  "I'm convinced the big enterprise integrators can sell anything.   A data mining project needs to be very well defined."