Demystifying Predictive Analytics and Modeling for Fundraising

Demystifying Predictive Analytics and Modeling for Fundraising

November 6, 2020
PV Bóccasam

When it comes to fundraising, your most valuable asset is your time. Predictive analytics and modeling help determine which prospective donors to focus on so you can secure more gifts faster. However, the initial complexity of the process can seem overwhelming for those unfamiliar with it. 

This guide describes what predictive modeling is, why it’s important, and how to get started. It’s based on a workshop led by Eric White, a Senior Principal Consultant with WealthEngine, during the recent WE Prosper Summit. 

Click here for a replay of WE Prosper Summit, a virtual conference exploring trends and best practices for finding high-potential donors in the COVID economy. 

What is Predictive Modeling?

“Predictive modeling is [the] analysis that allows organizations to evolve from a subjective approach to fundraising to a data-driven approach,” explains White. Instead of ringing up every prospective donor and seeing who bites, modeling offers fundraisers a clearer picture of who has the highest probability of making a major gift.

“Modeling gives a numeric value to categorical data,” White notes. “The formula delivers a score for your constituents based on the desired behavior, such as making a major gift….A raw score is a number between 100 and 1000 with those near 1000 having the highest probability of, in this example, making a major gift.”

The 3 Things Needed for Predictive Modeling

  • Refine Your Question

“The first thing you need to recognize is a model can only answer one question,” observes White. “Are you interested in who’s most likely to give you a gift of $25,000 or more? Are you interested in someone making a monthly gift of at least $25 or more? Are you interested in finding individuals that will include you in their estate plan? You want to refine the question.”

  • Clean Your Data

Improperly selected or entered data can derail your entire model. Just ask the crew of Emirates Flight 407. In 2009, during a flight from Melbourne to Dubai, they accidentally entered the weight of the plane and 270 passengers as 262 metric tons. 

“Its actual calculated weight was 362 metric tons,” notes White. “This is the equivalent of not calculating the weight of 20 African elephants stored in the belly of the plane.”

While the plane eventually got off the ground, it was almost the worst civil air disaster in Australia’s history. The moral of the story: you have to be meticulous with your data to get the desired results.

  • Choose the Right Variables

Predictive modeling involves plugging data into a formula. Each part of the formula is known as an independent variable, or the unique traits of the individuals being screened.

“For predictive models in the fundraising space, you need eight donation variables,” advises White. Those include:

  • First gift date
  • First gift amount
  • Largest gift date
  • Largest gift amount
  • Last gift date
  • Last gift amount
  • Total number of gifts
  • Total amount given

Depending on the type of giving you’re focusing on, you may need to include additional variables. For example, the age of donors is important to know when modeling planned giving over several years. You also need the data of at least 200 individuals in order to have a stable model. 

The 5 Steps of the Modeling Process

  • Model Design

White and his team spend time discussing what the client wants to achieve with predictive modeling. Knowing the end goal allows them to design a more accurate model.

  • Data Preparation

This is the process of gathering, cleaning, and inputting all the data into the model. “Paying the price on the front end and cleaning the data positions us on the kinds of returns on investment we’re all looking for,” advises White.

  • Model Building

“That’s when the data scientists come in and they start reviewing and working with the data,” explains White. During this step, they determine which of the included variables are most important for the model.

  • Performance Analysis

White’s team carefully checks that the model is accurate and effective. “We want to make sure that it’s a good fit, that it’s a robust and viable model,” notes White.

  • Modeling Scores

At the end of the process, White explains, “You get two things, a raw score and a major gift decile, along with best practices and continuing consulting issues around the best way to implement the scores and ratings.” A raw score is a number between 100 and 1000, while a decile places prospects into groups. 

“Let’s say you give us 100,000 records,” explains White. “We’ll score everyone from the first person to person 100,000 with a raw score and then we’ll rank order them and divide them up into 10 equal deciles. The individuals in decile one are the ones that scored in the 90th percentile or the highest level of the model.” 

Predictive Modeling Drives Results

To cap off the talk, White shared a case study of a university his team worked with.

“They gave us over 100,000 records for us to screen,” he recalls, “and we built them [a model] that predicted individuals most likely to give at least $100,000.” 

The result was a list of 220 individuals that were likely to give at least $100,000 to the organization. The major gift officers selected 10 random people from the list who had never given to the university and called them. 

“All 10 took the call,” notes White. “Out of those 10, three said, “Let’s talk.” Out of those three when they met with them, one of them said, “What took you so long to call?” They ended up working with this individual who had never given to the university.”

The story doesn’t end there. The organization put together a proposal asking for $150,000. The donor ended up giving even more than that. The university’s partnership with WealthEngine paid for itself several times over. 

Want to learn more about how WealthEngine can help you achieve similar results with predictive modeling? Get in touch at info@wealthengine.com or click here to schedule a free demo.