Despite the popularity of this approach, there are significant limitations with manual data mining.

• Too few techniques are used. Traditional approaches to data mining use only one or two techniques to attempt to find the pattern in the data that will be predictive of a given outcome.

• Too few permutations are tried. Often the algorithm is correctly chosen but not enough permutations are run to find the best model (e.g., Permutations include sub-algorithms, iteration criteria, entry/removal criteria, etc.).

• Too few data transformations are tested. Often the algorithm is correctly chosen, but the data do not fit the requirements of the algorithm (e.g., linearity).

• So many predictors, so little time. There are more predictors than we have the time to test. (especially since we must test them in combination with other predictors)

• Certain predictors "mask" the effect of other predictors. A technician would say that the "structural equation is not parsimonious."

• It's unproductive. Even an expert modeler can only produce 5 to 10 optimized models per week. With only a handful of models built, it is unlikely that the highest performing model has been discovered.

• There are four other critical, higher level data mining issues that consistently prevent cross-sell and up-sell results from producing optimal outcomes. We have developed our Ultra-High Performance Data Mining^SM (UHP-DM) to address these limitations. For more on the UHP-DM^SM inspect the Proprietary Approaches section under the Company menu of our site.