Five Keys to Maximizing Your Predictive Modeling Results
Guest Post by Erin Moran, Partner, The Solas Group
Thanks to wealth screening tools like iWave, it is easier than ever to get information about your prospects’ giving capacity. In this age of comprehensive, on-demand wealth screenings, information that used to take several days to gather on a single prospect is accessible on all the individuals in your database in a matter of seconds. The availability of information on wealth and philanthropy has made quick work of understanding financial capacity, be it of a single prospect, a whole portfolio, or an entire campaign.
Unfortunately, when it comes to understanding a prospect, giving capacity is only half the story. The other half—the prospect’s willingness to support your organization—is also necessary, particularly as you work to prioritize the many good prospects that are being surfaced. And because qualifying every individual in person is unrealistic for most development offices, organizations need some way of predicting which individuals will be most responsive to a solicitation. This is the role predictive modeling plays in your development toolkit.
Below are five keys that will help you succeed with any modeling project.
1. Understand that Predictive Models are Not All the Same
When it comes to models, there are many options. You’ll want to start by knowing what behavior you’re trying to predict. A model to predict the likelihood of making an annual fund contribution may be entirely different from a model to predict major giving. And, of course, while we think of “planned giving” as a single category, it is itself highly diverse. A person who may be likely to make a bequest to your organization isn’t necessarily more likely to give through a unitrust. Whether you’re building models on your own or working with a vendor to create them, it’s important to understand what the models are intended to predict.
Another consideration is the data informing the model. Some models use data gathered from the general population, predicting whether your constituents are more likely to give in a certain way based on aggregated data on people who aren’t your constituents. For example, such models may predict that individuals over the age of 50 may, in general, be more likely to respond favorably to a bequest solicitation, so your constituents over age 50 will be rated accordingly.
Custom models are developed from your own internal data, built to predict which attributes among your own constituents correlate with certain giving behaviors at your institution. These models may end up being similar to those based on the general population, but they are more precise because they allow for variances unique to your organization.
Done well, custom models are the gold standard of predictive modeling. However, custom models are also highly complex, so make sure whoever builds your model has a deep knowledge of both modeling and fundraising. Your modeling provider should be able to clearly explain how they isolate dependent versus independent variables so that they (and you) can have confidence that the model is truly predictive of the giving behavior you’re looking for.
2. Conduct a Comprehensive Wealth Screening First
The more accurate and complete your data is, the better your model will be. By conducting a comprehensive wealth screening prior to modeling your database, you can bring all that new wealth information to bear when you develop your model. It may be that some of the information you learn, such as having specific types of assets or making charitable gifts to specific other organizations, may be predictive of desired giving behaviors at your institution.
Beyond just wealth screening, you will want to be confident that your data is in good condition. Conduct all the routine data appends (addresses, phone numbers, etc.) before you model. Make sure you don’t have any data entry projects you’ve been postponing. If, for example, there is an event attendee list that hasn’t made its way into your database, you’ll want to add it in case it helps to predict giving behavior.
3. Get Buy-In
Fundraisers always love getting strong leads, but a good predictive model will produce so much new information at once that they need time to prepare for how they’ll manage the results. You can help major gift officers by reviewing your organization’s upcoming events and travel plans to recommend ways of incorporating their new leads into mailing and invitation lists. You will also want to speak with the heads of annual fund and planned giving to make sure they are aware of how the data will be delivered. Depending on the nature of your modeling project, you may receive scores for annual giving and planned giving that can be directly incorporated into their ongoing telefund and mass marketing projects. Speaking of mass marketing, make sure you tell your communications colleagues about the project ahead of time. They may be able to leverage upcoming mass communications to inform and engage individuals with strong modeling scores.
When talking with your colleagues about the model, it’s important to manage expectations. Modeling is about probabilities, not certainties. Some of the individuals the model identifies as being strong prospects will not be as responsive as you might hope. Other prospects you already know to be well-qualified may not receive good scores because they may have attributes that make them less “typical” of a prospect to your organization. Neither of these facts means that you have a bad model. Your model should save fundraisers time and point them in the right direction, but anomalies will occur so you and your colleagues should expect them.
4. Make a Plan
After you conduct a comprehensive wealth screening and develop predictive models, you will have a lot of data analysis to accomplish. The first thing you will want to do is sort and count the records by category. How many individuals can you find, for example, with high likelihood and capacity to make a major gift who are unassigned? How many individuals have a high likelihood to contribute to the annual fund that have not made a gift in more than one, two, or three years?
From a major gift standpoint, looking for donors with the highest possible wealth and likelihood scores is an obvious step. Less obvious, but also important, is to look at the individuals with high wealth and slightly lower likelihood, or high likelihood and slightly lower wealth. Are there events that might be particularly appealing to the wealthier, less likely donors to bring them closer to your organization? Are there leadership annual fund strategies you might employ to keep your high-likelihood, less wealthy donors “within the fold”? Always remember that the data you’re analyzing is a snapshot of a moment in time. Life circumstances can influence any individual in such a way that they may become more or less likely to contribute, or indeed more or less wealthy. If individuals build their wealth, or if they decide to make your organization a bigger philanthropic priority, you will want to be able to take advantage of those fortunate circumstances.
5. Track Your Progress
Once you’ve gone through the process of building a custom predictive model, you will want to refresh the data regularly and re-develop a custom model occasionally. Because predictive modeling is an ongoing process, it’s important to evaluate how your model performed. The best way to do this is to develop and maintain a system for tracking the records that your model helped identify so you can measure the success of your effort. How many new major gift prospects did the model identify? How many were qualified and assigned? How many made a major gift?
Such measurements can also help fundraising managers make sure that their staff members are taking advantage of new leads. How many prospects did the model identify that were not contacted in the last three years? Over the course of a year, how many of those prospects were your development staff members able to successfully contact? How many non-donors became donors, and at what level? If you can measure the positive impact the model had on your fundraising results, you will have a strong case for continuing to conduct modeling in the future.
Predictive modeling represents a worthwhile investment of time and resources. With a little planning, you can ensure that that investment pays off and is able to strengthen your entire fundraising program.
Interested in learning more about predictive modeling? Check out the recording to this awesome webinar here!
Stay Up-to-date on Fundraising News and Resources