by Scott Rosenkrans
Head of AI an Growth, EverTrue
Every fundraising team has run a wealth screen. Fewer have asked the question that actually matters: not who can give, but who's ready to.
That distinction is where most prospect research programs quietly stall. A wealth screen hands you a list ranked by net worth, real estate, and business affiliations. It's a real starting point. But it doesn't tell you whether that constituent has ever made a charitable gift, what causes they care about, or how likely they are to respond when your organization actually makes the ask. Capacity tells you the ceiling. It doesn't tell you who's close to reaching it.
The programs pulling ahead right now have made a shift: from capacity-first research to behavior- and intent-first research. The tool behind that shift is predictive modeling, and it's more accessible than most teams assume. It doesn't require a data science team, a six-figure build, or a multi-year implementation. It requires the right data, the right model, and a clear sense of what your program needs answered.
The Four Questions Worth Asking
Here's a fast test for your current research infrastructure. Can your team answer these four questions with a reliable, data-driven yes, not a guess?
- Who in our non-donor pool is most likely to make a first gift?
- Which first-time donors are we at risk of losing before they give again?
- Who in our current donor base has real room to give significantly more?
- Who should we be cultivating for a recurring giving program?
If those feel hard to answer with confidence, that's not a failure of research instinct. It's a tooling gap. Most teams are working from capacity scores, tribal knowledge, and whatever the CRM happens to surface that day.
This matters more than portfolio size. Gift officers are already carrying more names than they can meaningfully work. The challenge was never generating more activity. It's making sure the activity that already exists is pointed at the right people. A researcher who walks into a portfolio review with "these 40 names scored highest on our upgrade model, and here's why" is having a fundamentally different conversation than one who sorted by capacity and flagged the top quartile.
National Models vs. Custom Models
Predictive modeling sounds like it belongs to enterprise teams with in-house data scientists. For most organizations, it isn't.
National models are built by a vendor using giving behavior across anonymized data from thousands of organizations, then applied to your file against those broader patterns. No custom data science work required, and the lift over capacity-only screening is real.
Custom models are built from your organization's own giving and engagement history enriched with external philanthropic data. They reflect your donors specifically, and the accuracy ceiling is higher, but so are the data, financial and time investments.
For most teams, particularly those new to predictive modeling, national models are the right place to start: accessible, fast to deploy, and a meaningful upgrade in prioritization without the complexity of a custom build. A file with four years of clean gift transactions and a constituent list is enough to run most national models. You don't need perfect data. You need enough of it.
What This Looks Like in Practice
A global museum had segmented its year-end appeal into two audiences, visitors and concert attendees, for years. The team wondered whether that split reflected how donors actually engaged, or whether it was just an assumption no one had tested. They built a predictive model against four years of appeal data, scored roughly 25,000 constituents, and surfaced donors already in their database who'd never been part of a prior appeal, then unified the messaging around a single story instead of two.
A two-person foundation supporting a public school district had a goal that usually comes with a research team attached: shifting from event-driven annual giving to a multi-million dollar endowment. By layering giving history, engagement, and capacity signals onto their existing list, they found where the real opportunity already lived, small past gifts included, and walked into their board with a confident silent-phase strategy. No new hire required.
The common thread in both stories isn't "we found donors we never knew existed." It's "we found donors who were already there. We just didn't know to call them."
Where to Start
Don't try to implement every model at once. Pick the goal in front of you right now, an event, a lapsing donor problem, a leadership giving pipeline, and start with the model built for it. Get comfortable with the score, build it into the workflow, and let the next question tell you what to add.
Capacity tells you the ceiling. Predictive modeling tells you who's ready to reach it. The organizations doing both well aren't guessing anymore, they're making decisions based on evidence.
Curious what predictive modeling could surface in your own database? Request a demo of EverTrue and see which of your constituents are already ready to give, you just don't know it yet.
Want the full breakdown, including the four questions, the model-by-model decision framework, and a checklist for evaluating predictive modeling vendors? Download the complete guide, Beyond the Wealth Screen.
Scott Rosenkrans is Head of AI Growth at EverTrue, where he helps nonprofits apply AI and analytics to strengthen donor engagement and drive mission impact. With nearly a decade of experience designing predictive and generative AI products for the nonprofit sector, Scott specializes in translating complex technology into strategy, training, and services that deepen—not replace—human relationships. He’s a Certified AI Governance Professional (AIGP), co-host of the Fundraising.AI Podcast, and co-author of Nonprofit AI: A Comprehensive Guide to Implementing Artificial Intelligence for Social Good.
