Partial unknown system: If a model of the system that produces the data is well
known, there isn't any need for Data Mining because nothing is hidden and all
variables are in some way predictable. That isn't the case of e-commerce, because of
the effects of human behavior; whether, and political decisions, among others.

In those cases there will be knowledge of part of the system and there will be a
component apparently of random nature. Under certain circumstances, from a large
amount of data associated to the system, there is a possibility to find new
aspects previously unknown of the model.

Huge amount of data: Large streams of data compensate the limitations of an
incomplete model. That is particularly true when Neural Networks and other
adaptive techniques are used. In theses cases, enough data for training and
verification will be needed.

Powerful Hardware and software: Most of the tools present in Data Mining are
based on the use of massive computation, therefore convenient hardware and
efficient software increase the performance of the process, that some times has to
deal with production of data at levels of Gbyte/hour.