The current fundraising climate is challenging. Between rising costs and an uncertain economic outlook, organizations seek ways to contact the right constituents at the right time and ask for the right amount. Machine learning models provide an increasingly effective method for predicting the most responsive donors compared to traditional RFM segmentation approaches.
According to Lee Gartley, Head of the MiLo Intelligence Machine Learning Team at ROI Solutions, “Much of response modeling typically focuses on marginal populations, aiming to identify the small percentage of records that can be contacted successfully. Examples include reactivating lapsed donors or attempting to convert one-time donors to sustainers. With recent increases in mailing and production costs, many organizations are also looking to improve targeting in more extensive active donor campaigns. Here, the goal is to identify the small/medium percentage of donors that are more economically viable to include in a specific campaign.”
Postage costs have risen dramatically over the past few years. As The Nonprofit Alliance reported, nonprofit mailers have seen postage “increases between 15.7% and 19.6% in a mere 18-month period” click here to read the article. With postage as just one of the rising direct marketing costs, our clients are looking for new ways to maximize returns with shrinking mailing budgets.
Most organizations hyper-focus on their active annual donors, but as costs go up, how can you be sure you are mailing the most responsive donors? With active donor file modeling, you can rank all active donors on the likelihood of giving and improve targeting while reducing overall mailing costs.
As an illustration, in the table below, which is based on an actual campaign, the organization could drop classes 9 and 10, which represent 20% of the contacts but only 2.6% of the gifts, or even include class 8, which would drop 30% of the contacts but only 6.4% of the gifts.
Adding columns for cumulative contacts, responses, and revenue illustrates this more clearly. For example, 50% of the contacts (all records through MiLo class 5) represent 84% of the responses and 88% of the total revenue.
As discussed in our previous update, MiLo Intelligence Approach to Machine Learning Modeling, three of the most important requirements for response modeling adoption are:
- The availability of high-quality and consistent data.
- A well-defined marketing process to maximize the modeling scores.
- Minimal upfront investment or risk.
On the data front, the data for active donors is most in line with current best practices for data collection within your organization. Model predictability always improves when you mirror ongoing solicitation strategies for model development.
On the process side, active donor model scores are easy to leverage with an audience contacted frequently. A potential risk with an active donor file is over-solicitation. Our clients avoid this in different ways. One way is through the selection process, where specific names are intentionally excluded based on prior contacts. Another method is to include a frequency feature in the model build process, which will impact predictions based on how many contacts each constituent has recently received. The model should help exclude the constituents negatively impacted by over-solicitation while including those most likely to respond.
Our philosophy on the Milo Intelligence team is to allow our clients to test new ideas and models with minimal upfront investment or risk. We have seen organizations struggle, particularly in this economy, when a new model’s trial requires a material investment in dollars or time from the organization.
Machine learning can be a powerful tool for gaining insights into constituent behavior and fine-tuning solicitation strategies. Still, it is essential to use the models thoughtfully, based on your own experience and critical thinking.
Learn more about MiLo Intelligence from ROI Solutions and our approach to predictive machine learning models. Let’s talk to maximize the value of response modeling in a challenging economy.
