How to Develop an Analytics Project: Building a Model Score

Missy Garner, Director of Prospect Research
Rick Loveday, Prospect Research Analyst

Part 1: Concept Development

This post is the first in a four-part series that will outline how Clemson University’s Prospect Research Team developed a philanthropic model score.

This project began as a way to identify key philanthropic characteristics in our constituent base.  We wanted to know what types of attributes drive people to be philanthropic. This brought us to a specific group of our constituents who had donated at the principal gift level. We wanted to see if this subset exhibited any shared characteristics beyond the standard information contained in our CRM such as event participation, board membership, demographic data, etc. No matter how strong our data is, it only provides information about a constituents’ relationship with our institution. Our data won’t tell us how our constituents are interacting with other organizations or institutions. This drove us to look for outside resources that would fill this gap in our data.

We previewed several products and found that DonorSearch provided the largest breadth of philanthropic data. They have developed their own database to track specific philanthropic history. This allows them to identify key attributes such as donations to other organizations, political contributions, board or non-profit affiliations, and high-level business positions. While other companies such as Blackbaud and iWave track these data points as well, DonorSearch made these philanthropic indicators their main focus.

We decided to send a test batch of constituents as a proof of concept. We also needed to determine whether the Development team would embrace the results, since our previous models for determining a prospect’s value were primarily based on hard asset screenings. This new model would focus on where a person donates their money - not on how much money they have.

The original test group was made up of just our principal gift level donors. We wanted to work with a larger group, so we included alumni and current parents who, based on our Research Point wealth screenings, had estimated wealth of $1 million or greater. This provided us with an adequate sample size to hopefully apply some of these philanthropic indicators to our group. After signing a contract and a non-disclosure agreement, we determined what constituent information should be shared for this screening. In order to screen our group, DonorSearch, at minimum, needed constituent name and address, with the option to send other data points as we saw fit. We also included date of birth, spouse name, and employment information. Since this was our proof of concept we consciously chose not to send any specific giving information.

We sent the data off using their secure server and received our results within 24 hours. The file contained over 120 data points for us to digest.  We wanted to verify these philanthropic indicators fit with the data that was returned. In order to do this, we would need to pore through each and every data point. 

The next blog post will detail how we determined which data points we would use to build the model.