Machine Learning Advancements Power Digital Fundraising Growth

Ashwin Narayan, Principal Data Scientist, MissionWired

In 2023, we saw significant advancements in the machine learning space: ChatGPT seemed to be everywhere, dominating news headlines and making its way into countless dinner conversations. 

As a data scientist, I was incredibly excited to see the worlds I’d been deeply embedded in break into our everyday lives – and even more excited to bring these advances to MissionWired and, importantly, our partners. MissionWired has been on the forefront of artificial intelligence and machine learning and its application to fundraising, one of the reasons I was eager to work here. But as the machine learning field has evolved, we have, too.

Over the last year, we kept circling a set of questions, the core questions of the ML practitioner: How do you glean predictive power from the reams of data in The Digital Co-Op? How do you map the major advances in model power to your core use cases? And can we do so in ways that accelerate return on investment for our partners and help drive growth and revenue for their campaigns and causes?

We took a big swing, and no aspect of our models were out of bounds (as the whiteboards in our offices can attest). After rigorous testing and engineering, hypotheses confirmed then refuted, and more engineering, the answer to that last question is unequivocally yes.

I’m thrilled to share that we’ve developed a new modeling architecture that has improved The Digital Co-Op’s already best-in-class targeting capabilities. While we’ve consistently been improving our model since TDC’s launch in 2020, this new, innovative update is decidedly a leap forward: more accurate in finding your best donors and more efficient in crunching through data.

I’ll share more on the how the model has evolved and where we hope to go from here, but I first want to share some of the really compelling results this new approach is driving:

  • In the political sector, an in-cycle Senate campaign’s acquisition investment broke even 35% faster than with the old model and converted 22% more donors. A large committee achieved an almost 60% lift in revenue per name and 56% more donors. 
  • How about nonprofits? For an organization focused on gun violence prevention, the new architecture broke even 20% faster when compared with the old model and converted 17% more donors. For an international aid nonprofit, it was a 52% greater return and a 76% increase in donors converted.

In both cases, the new model will continue driving even stronger growth, because the hardest gift for an organization is almost always the first one.

How is this new model different?

We’ve iterated on our previously successful model, which adapted the then-state-of-the-art in image recognition, and updated it to a more sequence-based model following the evolution of the machine learning field. We can now use the full power of the sequences of behavioral data that make up our Digital Co-Op, in their full resolution, to predict who will most likely give to your organization – now and in the future. 

While the science behind our new models is truly innovative and unique, the main thing to know is that this new architecture allows us to find even more highly qualified donors for your mission.

Putting it to the test

After our team finished developing this new architecture, we had to see if it would drive the results we hypothesized. Through a series of back tests and field tests, across a broad range of representative clients, we monitored three key areas: conversions, revenue, and negative signals. We always want to drive big revenue gains for our clients, but we never do it at the expense of deliverability. That’s why negative signals are incredibly important to us, as we employ rigorous screening techniques to protect deliverability.

As you can see from the numbers above, the results have been truly amazing – and they’re going to get better every day. 

Endless possibilities ahead 

I can’t wait to see what the powerful new modeling capabilities can do for our clients’ missions – particularly as we look ahead to the election this year. As we’re always evolving and looking ahead, I’m eager to continue optimizing and iterating this model. It could be hugely impactful in answering more burning questions: Who are your next best advocates, monthly donors, clickers, petition signers? Who will respond best to an SMS, direct mail piece, online ad, or email? I’m confident the teams at MissionWired will be answering these questions – and more – soon. 

To be one of the first members to take advantage of our state-of-the-art upgrades to The Digital Co-Op, reach out to your co-op contact, or get in touch at