– A Data-Driven Approach
Non-profit organizations (NPOs) play a vital role in addressing many of the world’s most urgent challenges: from the consequences of natural disasters and armed conflicts to food insecurity and lack of access to healthcare. In the face of these global issues, public generosity has remained strong. In 2023, an estimated 4.3 billion people contributed time or money to help others, reflecting the global commitment to collective care and support. NPOs rely on this generosity to deliver essential services, but transforming goodwill into lasting impact depends on their ability to fundraise efficiently.
Fundraising is the financial backbone of most NPO operations, with individual donations making up a large share of funding. While digital channels are growing in importance, traditional outreach methods, especially mail campaigns, remain critical, particularly for reaching older donors who often contribute the most. Mail campaigns are trusted and personal, offering a powerful way to tell stories and build donor relationships. However, they also carry significant drawbacks: high costs and environmental impact. Thus, NPOs must not only raise funds, but do so intelligently, by reaching out to the right donors with the right message at the right time.
In collaboration with the national branch of a global non-profit operating in over 130 countries, we studied how mail campaigns are planned and executed. Our partner organization runs 13 campaigns annually, each focused on a specific theme such as education, healthcare, or well-being. These campaigns account for nearly half of the organization’s revenue and a substantial portion of its fundraising costs. Although the campaign schedule remains consistent each year, decisions about which donors to contact for each campaign are made according to heuristic strategies. Improving donor targeting presents a major opportunity to increase net donations and extend the organization’s impact.
To address this, we developed a data-driven targeting framework that integrates clustering techniques with multi-armed bandit algorithms to infer donor preferences from limited data. Traditional machine learning methods, such as collaborative filtering or contextual bandits, were unsuitable due to lack of historical data and unavailability of contextual features. Instead, we assume that donors can be characterized by a finite number of types, and our algorithm learns both the types’ response profiles and donor-to-type assignment over time. Our method allows for more accurate targeting with fewer interactions, which is especially valuable given the short average lifespan of donor engagements.
We tested our framework on five years of data from over 1.5 million donors. Our results show that our method can increase net donations by more than 10% while significantly reducing unsuccessful contacts. These gains translate into higher fundraising efficiency, lower environmental impact, and improved donor experience. Ultimately, this project shows how advanced analytics can help NPOs turn limited resources into greater societal impact—supporting smarter, more sustainable fundraising in an increasingly complex world
