| THE DATAMART STRUCTURE
PROBLEM: In most environments, the datamart record structure is organized
at the household or account level, not at the customer level. |
THE DATAMART COMPLEXITY
PROBLEM: High performance predictive models typically end up using
only 1-5 percent of the available datamart fields in the final model.
How do we find the correct subset? |
THE PERMUTATION PROBLEM:
(The needle in the haystack problem): While millions of models could
be run to fully investigate the best solution…current manual
methods produce less than several dozen per week. |
THE CONDITIONING PROBLEM:
Invariably, any given model performs better with one segment than
another (i.e. they are better conditioned to perform with some customers
than with others). How can we arbitrate between all models to find
the best prediction overall? |
