This article by Seth Grimes caught my attention while at the Teradata Partners show - Limits on Prediction. In it Seth discusses what some of the limitations of predictive analytics might be, a topic much in mind here at the Teradata conference. Although there is a lot to what he says I think he misses a key approach to both maximize the value and sophistication of predictive analytics while minimizing the problems of complexity.
If the environment in which analysts build sophisticated models using these approaches, not suitable for the casual user, can embed the resulting models in an operational, transactional system then the casual user simply gets a "smarter" transaction. The fact that a math or stats wizard has used complex algorithms to build a prediction is invisible - the embedding of that prediction in an automated business decision let's regular folks take advantage of that sophistication witout having to be able to do it themselves.
EDM, in other words.