Key Takeaways:
- A fixed budget and an oversized voter or donor universe force a choice: spread contact evenly and waste it, or rank the universe and spend where it counts.
- The same person can rank differently as a next-best voter, next-best donor, and next-best volunteer. Scoring for one action and applying it to all three wastes the ranking.
- A ranked list only pays off if it reaches the team that acts on it. Turnout media, fundraising, and volunteer recruitment need different lists, routed to different teams.
You’ve got a challenge: You‘d like to contact every voter, donor, or volunteer on your list, but your budget is fixed. What you need is a way to reach the highest-propensity names on the list while adhering to the resources you have.
Why an Even Spread Wastes Budget
When you don’t have a ranked prioritization list, budget tends to be spread out one of two ways: either evenly, across the entire universe of voters, donors, or potential volunteers; or unevenly, by geography and based on where past performance was strong. Both approaches treat contacts as roughly interchangeable within a region or list segment.
As a result, low-propensity contacts are reached simply because they happened to be on the list, while people who are genuinely persuadable or ready to mobilize wind up being under-contacted because nothing separated them from everyone else. The gap here? You need a way to rank the data you have.
You need to start by thinking about your universe in a different way. It’s not a list of all the names to be contacted; rather, it’s a list of who’s eligible. By ranking it, you’ll know who to contact first, second, and not at all given the budget you have available.
Rank Your Voter or Donor List by Likelihood and Action Type
A vote, a donation, and volunteer work are three different actions, and the same person can score very differently across all three. Someone highly likely to vote may have no measurable propensity to donate, while an individual with strong donor propensity may have zero interest in knocking on doors.
Treating these as one score, or assuming a single “high propensity” ranking applies across all three actions, misallocates outreach. A next-best-voter model, a next-best-donor model, and a next-best-volunteer model need to be built and ranked separately, even when they’re drawing from the same underlying population and the same underlying attributes.
Now, we’ll go over the steps you need to take to build and rank your own models.
How to Build and Rank Your Own Next-Best Voter, Donor, or Volunteer Models
Step 1: Start with the outcome, not the model.
First, define what you mean by “high propensity.” If you’re looking at a list of voters, that could mean their likelihood of voting in the next election. If your list is comprised of donors, that may be their likelihood of giving within a specific window or over a certain amount. The key is to be specific with your definition. This will make it easier to determine whether your model got it right later on.
Step 2: Pull together your labeled data.
You’ll need a labeled set of data for your model to learn from. Specifically, this will require you to put together examples of people who took the action (ex. people who donated) and those who didn’t. Match this against the data you have on both groups, including past contact history, past giving or turnout records, demographic file data, and any survey or canvass data you’ve collected. The people who took the action are your positive examples. Everyone else in a comparable population is your negative example.
Step 3: Choose variables that plausibly predict the action.
Past behavior can be a strong predictor of a future behavior, but it’s not the only signal you should focus on. Demographic and geographic variables, engagement history (opens, clicks, event attendance), and any attitudinal or issue-based data you’ve gathered all add predictive value.
Step 4: Build a separate model for each action.
Whether you use a simple weighted scoring rubric or a more advanced logistic regression, you need to build and train separate models for the actions you’re looking at. In other words, do not use the same model for voting and donor propensity. Reusing one model’s output across more than one action will result in people being ranked incorrectly. The variables that predict voter turnout, for instance, don’t necessarily predict donating money.
Step 5: Score your full universe, not just people already in your file.
In addition to scoring people you have data on or have already contacted, extend your scoring to a broader file where possible. This might mean working with a vendor or public voter file to fill in the population beyond your existing contacts. Even though it’s an extra step, it’s important because it will help you surface high-propensity people you haven’t reached yet.
Step 6: Rank within each model’s output, then set contact thresholds based on budget.
Once each action has its own score, sort your universe from highest to lowest propensity for that action. From there, the ranking becomes a budget conversation: given the outreach dollars available, where does the marginal contact stop being worth the cost? That threshold should be set based on what your budget can actually cover.
Step 7: Validate against a holdout group before rolling out fully.
Before committing your full outreach budget to the ranked list, test it against a smaller holdout by contacting a sample of high- and low-ranked people and comparing actual conversion rates. If the model is working, the high-ranked group should convert at a meaningfully higher rate. If it isn’t, you’ll want to revisit your variables or your labeled data before scaling spend against the ranking.
Step 8: Refresh the scores on a cadence that matches your campaign timeline.
Propensity isn’t static. New contact history, new donations, and new engagement data should feed back into the model regularly, especially as a campaign moves through different phases. A model that was accurate three months ago may be stale by the time you’re making final outreach decisions.
An Alternative to Building and Scoring Your Own Models
There are a couple of gaps here you may have already spotted. The first is time: Building next-best-voter, donor, and volunteer models in-house can take anywhere from three months to nearly a year. Few political campaigns and advocacy organizations have that long to wait. Then there’s also the question of ability to reach out to next-best prospects who are outside the list you currently have. That isn’t possible to do with limited, traditional data.
Resonate gives organizations a better, faster path to the same outcome. Our predictive modeling will provide you with ranked, individual-level propensity scores across voting, giving, and volunteering, delivered without the months of data assembly, model building, and validation that an in-house build requires. And because our political intelligence draws on a foundation of individual-level attributes already built and continuously refreshed at scale, organizations and campaigns don’t need to acquire an expanded file, hire a data science team, or wait for a model to be validated before they can act.
That speed to insight matters most when campaign timelines are compressed and budget decisions can’t wait for a months-long build. Instead of standing up new infrastructure, teams get a ranked list they can route directly to fundraising, turnout media, or volunteer recruitment right away, with scores that stay current as the campaign moves rather than growing stale between refreshes.
Ready to Direct Budget Where It Will Make an Impact?
To get started finding your next-best voter, donor, or volunteer, schedule a consultation with a Resonate data expert today.
Frequently Asked Questions
What’s a target universe?
A target universe lists everyone who is plausibly eligible or persuadable based on a certain action or belief.
What is a next-best-voter model?
A next-best-voter model ranks a target universe by likelihood of action so limited budget goes to the highest-propensity contacts first.
Can one score be used for voters, donors, and volunteers?
Not reliably. The same person often ranks differently across these three actions, so each needs its own model and its own ranked list.
How often do these scores need to be refreshed?
Propensity shifts as a campaign progresses and new information becomes available, so scores need to update on a cadence fast enough to keep pace with outreach decisions.
Do we need our own data science team to build a next-best-voter or -donor model?
No. Resonate’s predictive modeling generates the ranked scores directly so teams can act on a prioritized list without standing up or maintaining a model in-house.
