There was a great article in Predictive Analytics Times recently by my friend Dean Abbott – A Good Business Objective Beats a Good Algorithm. Dean, like me, talks about the importance of the “three legged stool” of business, analytics and IT. But it was the title that particularly struck me. As Dean says, it’s easy to “solve the wrong problem exceedingly well” with a good algorithm so it’s vital that you have a clear sense of what the business objective really is. This means understanding the desired outcome – what do you want to improve – as well as any conflicting objectives that must be traded off and the business context within which you hope to achieve this goal (what CRISP-DM calls “business understanding”). The question is how to model and describe this?
Here at Decision Management Solutions we are increasingly using decision modeling to express this business understanding. After all the purpose of predictive analytics is to improve the accuracy, precision, profitability of decision-making. It makes sense, therefore, that one should understand the decision-making one is trying to influence! By modeling the decision-making involved it becomes clear where the proposed predictive analytic fits in that decision approach. It is also easy to link these decisions to the performance metrics and objectives they influence, the business processes of which they are part and the various organizations who care. All this makes for a strong, well-defined business context and a clearly defined objective for a predictive analytic project.
You can read more on our approach by downloading this white paper on Decision Modeling for Predictive Analytic Projects. If you are interested in this approach, check out our decision modeling tool DecisionsFirst Modeler.