Continue in 2 seconds

ASC Builds Robust Predictive Models with ModelQuest Expert

Published
  • January 01 1998, 1:00am EST

PLATFORMS: ModelQuest Expert is run on PentiumPro 166 MHz with 32MB of RAM under NT 4.0.

BACKGROUND: ASC Database Marketing provides data warehousing services to support their clients' prospect marketing databases. The data warehouse is the basis for the selection of names and addresses for direct marketing activities. To achieve cost-efficient marketing activities, ASC develops and implements predictive models designed to optimize the efficiency aspects of these marketing activities, such as response or profitability per response.

PROBLEM SOLVED: Marketing efficiency optimization models have a relatively short life span (perhaps one to two years) as the marketing conditions change often. Also, models which are specific to a marketing stimulus perform better than models which optimize over differing marketing stimuli (i.e., the more models there are in place, the better the marketing results can be optimized). ASC Database Marketing needed a quick and reliable way to build robust predictive models. After working for several years with neural network technology-based software and hardware tools, ASC Database Marketing tested AbTech's ModelQuest Expert tool. It significantly shortened the development of predictive functions and that predictive functions are always validated when used with actual marketing activities.

PRODUCT FUNCTIONALITY: The ModelQuest Expert version of the product now includes the functionality of several searching strategies for developing the predictive equation at the GUI level. So far, only the CPM optimization and Error Network strategies have been utilized to develop models. ASC Database Marketing has found the Error Network strategy to increase r-square from 20-50 percent. The network-on-networks search strategy will be tested in the near future.

STRENGTHS: The reliability of the predictive functions in real-life applications and the speed at which such functions can be developed are the strengths of Model Quest.

WEAKNESSES: While the product allows for transformation of the predictor variables, it would be very tedious to arrive at the most appropriate transformation of a predictor variable within ModelQuest. ASC Database Marketing pre-processes the training sets predictor variables to arrive at the best transformation and then proceeds to supply ModelQuest with the transformed data.

SELECTION CRITERIA: When switching from statistical modeling techniques to neural network technology, ASC Database Marketing experienced an improvement in its direct marketing results predictions. However, the process of training neural networks was only slightly less tedious than statistical methods. With ModelQuest, ASC Database Marketing has found a fast and easy method to develop reliable predictive models.

DELIVERABLES: For ASC Database Marketing the most desired deliverable is the C language code which represents the predictive equation. This equation is then incorporated into the data warehouse for calculating predicted values, which are then used for selection criteria from the data warehouse. The training and validation reports are useful reference documents for understanding and comparing different model performances.

VENDOR SUPPORT: There was never a need for frequent vendor support. Initially, I had some general questions about setting up the proper positive and negative outcomes and how to prepare the data. AbTech has people on staff who understand these issues and guide you in applying their product to your particular situation.

DOCUMENTATION: The documentation is brief but explains the underlying workings in a modeler's language that laymen can understand. ModelQuest is very intuitive and the on-line help seldom requires you to go back to the printed manual.

Register or login for access to this item and much more

All Information Management content is archived after seven days.

Community members receive:
  • All recent and archived articles
  • Conference offers and updates
  • A full menu of enewsletter options
  • Web seminars, white papers, ebooks

Don't have an account? Register for Free Unlimited Access