FREE DM Review Site Registration!
Sign-up today and access DM Review on the Web!

Your FREE registration entitles you to:
FREE email newsletters

FREE access to all DM Review content

FREE access to web seminars, resource portals, our white paper library and more!

   

Bank of Montreal Improves Analytical Insights with KnowledgeSTUDIO

Angoss Software Corporation - KnowledgeSTUDIO

BACKGROUND: Bank of Montreal, Canada's first bank, is a highly diversified financial services institution that ranks as one of the 10 largest banks in North America with average assets of $205 billion. The bank's group of companies include Nesbitt Burns, one of Canada's largest full-service investment firms; Chicago-based Harris Bank, a major U.S. Midwest financial institution; and mbanx, the first North American-wide virtual banking unit. Bank of Montreal has an equity position in, and an alliance with, Grupo Financiero Bancomer, the leading Mexican financial institution.

PLATFORMS: NT workstations, Microsoft Windows 95.

PROBLEM SOLVED: KnowledgeSTUDIO is our data mining frontrunner for desktop analytics processing. In the past four years, we have been deploying this software and its predecessor (KnowledgeSEEKER) to analytical groups all over the bank in an effort to promote data mining and improve analytical insights in the bank data. These groups include credit risk, marketing, treasury, client contact center, credit card decision support, different product groups and management information. The software has been used for customer segmentation, cross-selling models, campaign preparation, predictions of defaulting on mortgage payments, analysis of credit risk, etc. Essentially ANGOSS' decision tree is used as the initial step in most analytical projects even where larger volumes of data are later processed using different technologies. KnowledgeSTUDIO is also used for identifying nontrivial data quality/discrepancy problems as an initial step of our data transformation processes.

PRODUCT FUNCTIONALITY: ANGOSS brought CHAID algorithms to perfection. Absolute ease of use is combined with great performance and accuracy. The data import function is very good allowing for a smooth import from a variety of sources and providing graphical insights in single variables. Generation of different pseudo codes simplifies promotion of models to production.

STRENGTHS: The product's strengths include excellent performance of CHAID engine together with interactive exploratory opportunity given to users and ease of use, even for novice and non-expert users. ANGOSS' support for PMML (XML-based Predictive Model Markup Language) is in line with our need to share models between different applications and to catalog models in our Data Mining Metadata Repository. Relative low costs allowed for mass deployment of Data Mining capabilities throughout the organization.

WEAKNESSES: The JAVA pseudo code is limited to the decision tree (CHAID) and is rule centric. It requires manual modifications prior to moving to production. We would like to see support for AIX.

SELECTION CRITERIA: The major criteria are:

  1. Intuitive and well- performing decision tree analysis.
  2. Was not limited to one algorithm (i.e., inclusion ofneural nets) and open for easy addition of other algorithms.
  3. Ease of data imports from variety of sources and rule generation.
  4. Cost-efficient solution allowing for mass deployment across the organization.

DELIVERABLES: Models identifying propensity to buy certain products, segmenting customers on preferred channel usage, predicting customers defaulting on mortgage payments, etc., were developed recently. Across our client group hundreds of models exists, a number of which generated rules that were promoted to production.

VENDOR SUPPORT: Pre- and post-sale support was always excellent. Not only does ANGOSS consistently react promptly to our needs, but also proactively gets involved in our projects, pilots and presentations to our clients. They also frequently offer very good small group training to users interested in learning more about algorithms, methodologies and data mining life cycles.

DOCUMENTATION: Documentation is complete and easy to understand even by business analyst users without statistical background. The easy-to-follow tutorial sets the majority of users on a course of action without any additional help.


Dr. Jan Mrazek is president of Adastra Corporation, a vendor-independent data warehousing consultancy with 170 consultants in North America and Europe. You can reach Mrazek at jan.mrazek@adastracorp.com.

For more information on related topics, visit the following channels:



Industry Vendors