Asking a donor for the right amount of money is a crucial piece of the fundraising puzzle. Ask for too much, and it could anger a prospect. Ask for too little, and it leaves money on the table. It often feels like a delicate dance or a guessing game, but statistics and machine learning (ML) can lend the process both science and accuracy. With recommended algorithmic ask amounts, the time that a fundraiser would normally spend looking at spreadsheets and dashboards can be greatly diminished. By boiling down data into actionable insight, your fundraisers can spend more time focusing on the aspects of fundraising that require a human touch, and leave the numbers game to the machines.
Curating donor-specific ask amounts begins with exploratory analysis, like understanding your organization’s average gift size, standard deviation between gift amounts, average age of donors, etc. Diving into the data you have and knowing the quality of that data allows you to make predictions about your donors and their behavior—all existing features in your data set.
Once you have a good idea of what your data looks like, it’s time to put it into action. Machine Learning can both predict and refine outputs as it understands and analyzes data in real time—in this case, gift amount is the output. But, there are many different subfields within the field of Machine Learning, so which is appropriate here? The two main types of Machine Learning are regression and classification. Regression is employed for continuous output (example: how many dollars and cents should I ask someone for?) and classification is used for discrete output (example: is this donor someone who is more likely to give $50 or $5,000?). It’s important to define the problem you want to solve before you jump into the actual algorithms you plan to use.
In a previous blog post, I wrote about using an algorithm called the random forest to predict donor segments for people who have no giving history at your organization. This same approach can be applied to the problem as an example of a classification task. At Gravyty, we use our own proprietary ensemble of machine algorithms to output an ask amount estimate for each donor based on both the organization’s data and our supplemental data and analyses. As the organization uses Gravyty, the system learns from ongoing donor activity and improves itself, producing more accurate and timely ask amounts for the fundraiser to use in their solicitations.
However, in the nonprofit space, a common problem is actually having too much data to throw into an OLS regression. It’s not good practice to include too many variables that aren’t significant as this can often cause your output to be less robust. This problem is referred to as the “curse of dimensionality.” Check out this fun post on StackExchange to learn more.
Implementing ML to help your fundraisers spend less time on work that can be done by a computer and more time on work that can’t be is now not just an option, but an imperative. In an increasingly competitive fundraising landscape rife with high fundraiser turnover and clustered giving (90% of donations comes from the top 10% of donors), nonprofits need to use their gift officers’ human capital to build more and deeper relationships with donors. At Gravyty, we envision a not-far-off future where fundraisers are liberated from digging through their database and spreadsheets, and instead, spending 100% of their time with donors. Machine Learning ask amounts are a powerful step in that exciting direction.
by Evan Lyle and Lindsey Athanasiou