Predictive Donor Models
Predictive Donor Models
Most of us do not have endless resources to mail and/or meet with every prospective donor to our institution. I know that at UC Berkeley that was definitely the case where we had 500,000+ alumni, yet limited staff and resources to contact, engage, cultivate and solicit all of our prospective donors. Thus, strategically prioritizing resources to drive the greatest returns on investment (ROI) is essential for any institution's development program – and developing predictive donor models is the best solution to do so.
Many of us in the field of development and alumni relations do not really know what data mining and modeling is. We do know that large campaign consulting organizations such as Marts&Lundy, GG+A, Blackbaud, etc. can provide us ratings and screenings of our alumni, which are typically used as the basis to assess the overall potential capacity of your campus-wide, principal-gift-driven campaigns. However, we typically do not know the process used to develop these ratings/screenings, the internal and external data that went into them, nor what the final scoring mechanism (formula) is. In some sense, we are sold a mysterious “black box” that we must assume has magical powers to predict our major gift potential, and when it’s time to refresh the scores halfway into your campaign, you must pony up more funds to refresh those donor scores.
Per my previous blog post, The Founder of Predictive Donor Modeling (https://www.dejanvryconsulting.com/blog/the-founder-of-predictive-modeling), I ultimately believe that institutions should embrace and internalize data mining and modeling into their own staff/departments to create customized predictive donor models that reflect their institution’s unique alumni/donor base and specific business needs, such as driving growth in: 1) annual giving through new donor acquisition), 2) leadership ($1K+) giving through upgrades, higher renewals and new donor acquisition, and/or 3) planned giving through donor identification. These are three great examples of where customized predictive donor models have proven to be very powerful and successful tools to prioritize your marketing and fundraising campaigns to strategically grow your ROI. Furthermore, I believe that predictive donor models, particularly ones aimed at these three business objectives, should be equally (if not primarily) based upon a donor’s inclination to give to your particular institution versus just their capacity, which is the predominant predictor in major-principal giving models. Inclination, which is reflected by involvement and engagement, is often the best predictors of annual, leadership ($1K+) and planned giving.
Thus, I believe that more institutions should embrace the process of data mining and modeling into their own practices by training staff on the process of mining their data for insights and developing predictive donor models to prioritize and target their campaigns, resources and outreach efforts. And I believe institutions should own their predictive donor models – knowing exactly what the models consist of and having the ability to refresh and revise them whenever they wish at no additional cost.
For those that are interested in learning just a bit more about data mining and modeling, please read my Data Mining & Modeling 101 below. And if you are interested in learning more and/or contracting me to develop some custom predictive donor models for your institution to maximize your returns on investment (ROI), please do not hesitate to contact me to discuss your particular interests and needs. I love working with clients in this capacity, and love sharing my passion for leveraging data mining and modeling to strategically advance our alumni relations and development outcomes. Let’s talk! [E: ldejanvry@gmail.com M: 510-589-5944]
Data Mining & Modeling 101
Data Mining: Data mining is a process by which you sift through numerous potential predictors of your desired outcome for positive correlations. The process involves running numerous cross tabulations (crosstabs) where you look to identify and isolate variables that appear to have a positive correlation with your desired outcome – making a first annual gift, making a first $1K+ gift, or making a planned gift.
Here’s a simplified example of what it might look like. Let’s say your institution has a dues paying alumni association where people can either be: 1) Lifetime Members, 2) Current Annual Members, 3) Lapsed Annual Members, or 4) Non/Never Members, and let’s say your desired outcome is to find new annual donors. The crosstab that you would run to see whether alumni association membership may be a potential predictor of giving to your institution would show you the # and % of those four sub-populations that are currently donors and non-donors. As you can imagine and is no surprise, oftentimes alumni association membership is positively correlated with giving, and then the % of donors is highest among the Lifetime Members and lowest among the Non/Never Members, and a diminishing amount from Current to Lapsed Members. Thus, you might earmark Lifetime, Current, and Lapsed Members each as potential behaviors that will help us predict giving to your institution.
Now, imagine repeating this process for all of the potential predictors of giving in your database! Zip Codes, Job Titles, Marital Status, Degrees, etc. etc. etc. This is data mining. You are literally mining your data for potential positive correlations of the desired outcome you are looking to predict. And boy I will tell you, this process is very very enlightening about your data quality and collection processes, and thus I think all institutions can really benefit by having their own staff trained on how to mine their own data.
Predictive Modeling: Predictive modeling is also a process, but the process of modeling ends with a predictive donor model that can then be leveraged to revamp your resource allocation and programs to maximize your returns on investment (ROI). Predictive modeling entails running statistical regressions on all of the potential predictors that you identified and earmarked during your data mining process. As you run your regressions, you will find that some of your potential predictors will not be statistically significant as a predictor of your desired behavior, and thus should be discarded from your model. If you recall your dreaded statistics class, the statistical significance of a variable is determined by the T-Statistic, which is typically significant if it’s equal to +/- 1.67. As you fine tune your regression analysis, you will keep watch on your R-Squared, which represents the overall power of your predictive model – so the greater it is the stronger your model is becoming, and vice versa. While technically the predictive modeling process can continue forever as you try to refine and strengthen your model, typically you run out of potential variables and variable combinations to try and settle upon the strongest model that you can develop with your current data.
Predictive Donor Models: Your predictive Donor Model is the final outcome of your data mining and modeling processes, and using the coefficients of your remaining statistically significant predictors, you multiply them by 100 to convert them into weights, or scores. You then use this predictive donor model to score all of your alumni and donors, and then begin prioritizing your development and alumni relations outreach efforts to the highest scored individuals who have yet to exemplify your desired behavior/outcome (giving, giving at $1K+, making a planned gift). Typically your predictive donor model might look something like this:
Lifetime Members: +5
Current Members: +3
Lapsed Members: +1
Married/Divorced/Widowed: +3
Two Degrees: +5
Wealthy Zip: +5
Etc.
And your predictive donor model outcomes might look something like this:
Name Score Donor?
Sally 100 Yes
Joe 100 No
Sue 70 Yes
Jim 70 No
Shevan 50 Yes
Silvester 50 No
Johanna 30 Yes
Jester 30 No
Zita 10 Yes
Ziljian 10 No
Etc.
Leveraging Predictive Donor Models: Now, once you’ve created your predictive donor models, that is when the true fun begins! Your predictive donor models in and of itself will not guarantee your different results from what you’ve experienced in the past. That’s like banging your head on a wall again and expecting a different result from the previous 100+ times you’ve banged your head on the wall. But what your predictive donor model empowers you to do is prioritize your resources to maximize your returns on investment (ROI) by placing greater resources where the most likely of positive outcomes will come from, and less resources towards where the least likely of positive outcomes will come from. This can com in the form of greater funds allocated to targeted mailings or personal outreach by frontline fundraisers. The way I like to show clients the results of their predictive donor models and to begin the discussion of how they should leverage their predictive donor models to make more strategic resource allocation decision is as follows:
Total Scores # Donors # Non-Donors Strategies
80-100 10,000 1,000 Increase Acquisition Efforts
60-80 7,500 2,500 Increase Acquisition Efforts
40-60 5,000 5,000
20-40 2,500 7,500 Decrease Acquisition Efforts
0-20 1,000 10,000 Decrease Acquisition Efforts
Strategies: Assess Donor Engagement,
Retention & Stewardship
Efforts
Summary
I really hope this summary of data mining and modeling has intrigued and inspired you to think about how developing some predictive donors models at your institution might help empower you to more strategically direct your precious resources to grow your outcomes by maximizing your returns on investment (ROI). Please do not hesitate to reach out if you have any questions or if you’d like to discuss how I might be of assistance or how you might be able to manage such a process at your own institution.
Additional Readings