My colleague Darcy made a great suggestion yesterday about using the Fair Isaac Christmas card to illustrate the ways in which analytics can improve decision-making
Now, this is a purely rules-driven approach. We have used judgment and expertise to develop a set of rules that we then represent as a decision tree. Pretty standard stuff.
But what if I could predict the likelihood that a friend would give me a gift this year and roughly how much they would spend? Clearly whether they gave me one last year is a key factor in predicting this but there are other things that would impact it like my ongoing relationship with them, how they liked last year's gift etc etc. If I replaced "Bought me a gift last year" with "Will by me a gift this year" I might make a better decision (from my perspective, not necessarily from theirs). Similarly perhaps I could predict which cousins would pay me back and which ones would not rather than just considering who owes me money.
But what do I really want to do here? Well let's assume I want to optimize my gift and card giving so as to generate the best possible response - get as many gifts as possible without buying unnecessary gifts and without being embarrassed that I failed to give a gift to someone who gave me one. Maybe for kids I want to see what will cause good behavior next year not just reward last year's behavior. And so on. Well then I might build a model of how all of this relates (a decision model perhaps) that would allow me to come up with an optimal decision tree given my constraints (total budget for presents, willingness to risk missing someone, guilt with respect to spouse and kids).
These three steps are the classic EDM ones - automate a decision using expert judgment and business rules, then add predictive analytics to improve these rules and finally apply optimization technology to come up with the best rules for true EDM.