OK, I'll bite on this one - as a Data Strategist, I can't really avoid it! Prediction (predictive modelling, oracling, etc) is but an eye on the future, with vision that gets blurred. Modelling on social networks can be done (I've done it in a forensic space on email) - privacy is a constraint on that data. However insightful it may be, it's not what you can see (via prediction) that matters. It's what you do! If the value at risk is "this big" or the opportunity is " this many dollars", actions are needed to realise the opportunity and mitigate the risk. In modelling terms this is about testing your options (with an informative test framework), and costing the the options against their expected success, and assuming the future will be somewhat like the past (to keep the test findings valid). The inventive part includes coming up the options, and which customer attributes to include (including social network connections, contexts and attitudes, as well as the usual demographics and behaviour measures). Tom.

excellent bite!Id also add that insight -- as in motivation and understanding play a key role

Motivations and understandings are ideal insights. Sadly these "why" insights are like psychophysics - inside the head of the consumer. Marketing research tries to tell us what those insights are by asking the consumer (and then projecting the responses into the future). Modelling from customer or prospect data tries to relate behaviour with future actions, and avoids pretending it is answering "why" questions. Can we do better than this? Sometimes research data and customer/prospect data can be fused. Sometimes empathy is the best approach. Sometimes experience counts. Or persistence (like the hula hoops). I wish I had a better answer than this.

Fusion is the key. You know I bought a book -- now understand why. A gift? for my daughters school paper/ To read on a plane? each takes you to a different motivation.