In the back of my head I keep a list of predictive models that I would like to build someday. It just makes sense: I started out by creating a general model for predicting any kind of alumni giving to our institution, and now I create models targeted at specific types of giving (Planned Giving and Phonathon), and one specific type of behaviour (event attendance). From here, all kinds of possibilities suggest themselves.
In the area of annual giving, I’m very interested in the predictability of increasing pledge size year-over-year. The Phonathon model I created does a fantastic job of predicting who will participate (whether or not they’ve ever given before), and who will give the most. Of course this delights and amazes me, but these days I’m getting excited by how well the model predicts which of last year’s donors will make a larger pledge in the current year. (I’ve already written about this: “Predicting who will increase their pledge“.) I’m wondering if I tagged the known Increasers with an indicator variable, could I train a decent model on them, and do an even better job of predicting who might respond to a boost in target ask?
On the negative side, I’ve also wondered if it would make any sense to create a model for “likelihood to default on a pledge“. This would be taking a page from the risk-scoring models I imagine banks, insurance companies, and credit card companies must use to flag customers most likely to default on a loan, file a claim, or engage in fraud. Of course, the question is not “can it be done?”, but “what business problem would this solve?” What would we do if we knew which donors are most likely to fail to follow through on their pledge? Would we start sending out reminders earlier? Would we remind more often? Would we do follow-up calling? Those are questions I’d have to ask people who work in Annual Giving.
And finally, I’m intrigued by the idea of creating a “mid-range gift propensity” model. It’s my impression that prospects with capacity to give at the very high end of the Annual Giving scale, but yet who fall short of the Major Gift threshold, are an overlooked and untapped resource. The dollar values will differ from institution to institution, but I’m thinking of giving in the range of $10,000 as an outright gift, to perhaps $25,000 over five years. Now, no model I create is ever going to say anything about capacity. But I strongly suspect that there is a segment of the alumni population who feel motivated to give much more than they do, but who need to be asked differently. Some donors will respond generously to a call from a student, but frankly I don’t think the generic student call is going to cut it for most people in this category.
Again, creating such a model requires coordination with fundraisers over what is meant by “different asking”. Perhaps it involves staff, or maybe peer calling (a former classmate who volunteers to make calls?). I don’t know. But if I were to suggest an Annual Giving experiment with a high potential ROI, this would be it.
What’s YOUR fantasy predictive model?