February 20, 2013

Annual Giving Analytics (Part 2)

The previous post in this Annual Giving Analytics series focused on how analytics can be used to further the goals of Annual Giving, but how exactly is it done? This post will document and describe a few methods and approaches that can get you going on furthering your use of the analytics for annual giving at your institution.

Annual giving is primarily interested in three variables that can influence the dependent variable of how much a particular constituent is likely to give: Segment, Method, and Timing. Segment is the demographic segment that a donor is categorized as being a part of, method is the channel or means by which an individual donor makes gifts, and timing is exactly when they are asked, either during the calendar year or relative to specific institutional events.   

Building a predictive model to assess the optimum conditions for annual giving can help your program to understand a number of things, like which donors are the most likely to give and therefore who should be prioritized by the program, where ask amounts should be set for the maximum contribution, and how each solicitation channel like email, mail, and phone should be utilized. Good places to start when building a predictive model are who is likely to renew, at what levels should people be solicited, and who is likely to upgrade to leadership annual giving.

Let's go through the creation of these three models step-by-step:

Giving/Renewal Likelihood Model
-          Flag donors in your database that have a renewal history.
-          Gather data points such as demographics, activities, interests, and geography related to these donors.
-          Use logistic regression or decision trees to rank donors based on probabilities to renew.

Ask Amount Model
-          Use a linear regression model to predict the largest or most recent gift given.
-          If the resulting “predicted” dollar amount is higher than the most recent gift amount then increase the ask.
-          If the “predicted” amount is lower then leave the ask amount the same.    

Leadership Giving Likelihood Model
-          Derive a binary dependent variable for existing donors at the leadership level.
-          Use a binary prediction method like logistic regression or decision trees to compare independent factors.
-          Output will be predicted probability that can be ranked.

Models like these can be constructed for any number of donor behaviors that might be important to your institution like sustainer/recurring, catastrophe response, preferred giving channel, or cause or interest specific appeals. But of course, all of these models are best used together in order to maximize efficiency and effectiveness.   Results for the programs for which the models are used to boost efficiency should be considered holistically as one strategy and results should be captured not only by appeal but also by demographic segment. Tracking, testing, and reporting mechanisms must be in place prior to launch in order to accurately measure performance improvements.