This month our blog comes from Vicki Williams and Abigail Mann; Prospect Research Analysts from The University of South Carolina.

At the University of South Carolina, we are creating a predictive behavioral model using attributes assigned to our constituents to predict the outcomes of future solicitations and identify high quality prospects with a strong affinity for our university.  

Our model came about from a simple question:  How does affinity influence a constituent’s giving?  We asked ourselves this deceptively innocuous question in Fall 2013, and it has morphed into a research project that has gone through several developmental stages as we have met challenges that had to be overcome in order to create a model that would work for us and meet our needs.  Over a year later, we are ready to test our model, but reaching this phase demanded a lot of time and hard work.  It also required us to meet with colleagues outside of our prospect research shop.  We learned early on that we could not build a strong, actionable model without our colleagues’ specialized knowledge and expertise.

To begin with, we defined the term “affinity.”  For our purposes, affinity came to be known as engagement with the university through involvement.  In our CRM, attributes are assigned to constituents for event attendance, participation in athletics, memberships in clubs, service organizations, and more.  We knew attributes were going to be the critical piece to creating our model, because they would be our way to measure affinity.  We proceeded to measure every attribute through assigning it a weight on a scale from 1 to 5, with one being the lowest level of involvement and five being the highest. Through this process, we essentially quantified involvement. 

There are over 1,400 attributes in our CRM, and, believe it or not, we weighted every attribute for level of involvement.  As we’re sure you have already guessed, this was by no means an easy process.  Due to the sheer volume of the data set, there were, understandably, gaps in our knowledge regarding the significance of many attributes. Not only was it necessary to conduct individual research, but also to collaborate with our donor relations team and development officers from several divisions in order to make sure that as many attributes as possible were clearly defined and weighted accordingly. There were debates and marathon meetings on what attributes warranted a five, what attributes warranted a one, and what some attributes were doing in our CRM in the first place. This led us to request Information Systems to generate reports of constituents with particular attributes to examine trends in total giving and other involvement with the university.  One example is our athletics teams. Each of the university’s twelve teams has its own attribute and all were initially given the same weight. We began to wonder, however if this was using too broad a brushstroke.  This motivated us to request a report of constituents with these attributes.  We found staggering differences in engagement and giving between the various teams.  Surprisingly, we found that constituents who had played golf for the university were more involved and gave more frequently than constituents with other athletics attributes.  As a result, we adjusted our weights for these attributes according to the data.  

With the attributes finally weighted, we will be working with Information Systems to integrate the weights into our CRM to enable us to generate datasets that show constituents’ affinity scores.  In the next month, we will be prepared to test the strength of our predictive behavioral model by running linear and multivariate regressions of a sample population to see how our attributes, coded by their weight value, can predict giving.  We are in the middle of choosing our first sample population for our test run.  It has been our experience that constituents who were members of a Greek organization during their college careers were also members of other organizations with some even serving in leadership roles, and, as a result, have a total weight of attributes that shows significant affinity. We have also witnessed trends showing that as constituents advance in age, their giving increases. Should we use this sample population, our goal will be to see how much giving increases as constituents with a membership in a Greek organization and high affinity score start to age.  Our results may reveal to us that giving by constituents who were members of a Greek organization and have a high affinity score increases by thousands of dollars from when they were fifty to the age of sixty. The strength of this relationship will have the power to help us make predictions about future giving.  We will then be able to consider increasing our solicitation amounts of constituents who were members of a Greek organization and have a high affinity score who are entering the retirement stages of their lives all based on the data results of the regression that the model enabled us to run.      

Another added bonus to the model is that we will be able to use it to fill our pipeline through identifying prospects that may have otherwise gone unnoticed, because we did not screen them through our customized model to see their predicted giving based on their affinity.    

Knowing the strength of the relationship between affinity and giving will help us to understand our donors better, and understanding our donors is most certainly one of the guiding principles of prospect research.  Our hope is that we will have persuasive statistical evidence that supports our empirical observations that affinity can predict giving.  We feel very strongly that we have created a predictive behavioral model which will enable our leadership to make data-informed decisions about our donors and future solicitation amounts.

In a couple of months, we will be excited to share the results of our project with you when we have precise numbers explaining the relationship between affinity scores and giving.  Stay tuned for another post about the outcome of our first application of the model as well as tips and tricks for how you can “DIY” your very own model.