Most nonprofits we talk to describe their audience targeting the same way: send to everyone who's opened in the last 90 days. That works. It also leaves the real question unanswered: given this specific email, who on the list is most likely to care?
Until now, no tool could really answer it. Vector embedding models (the same tech that powers large language models) changed that.
We just published research with Civic Shout that puts a number on it. Across 79 million anonymized email sends, we found that subscribers have distinct, measurable content preferences, and that matching content to those preferences lifts opens by +5.8 points and clicks by +5.3 points.
These aren't simple issue tags. A subscriber who engages with corporate accountability, consumer protection, and workers' rights petitions isn't in three segments. They have a coherent preference pattern a model can identify even when a human can't label it.
Civic Shout sends one petition per day to their active list today, selected by a strong editorial team. We asked a different question: what if they produced several petition options per day and routed each subscriber to the one they're most likely to sign? Applied retroactively to twelve months of their actual sends, the model projects +18% more clicks and an additional 1.1 million petition signatures per year. Each subscriber still gets one petition. They just get the right one.
A live randomized controlled trial with Civic Shout is underway. Full case study with results to follow.
Also new this month: a resource for the people hiring consultants
If you've hired a consultant, you probably figured out how the engagement worked as you went. There's no playbook for the client side of the table. Until now. A group of consultants in our community, led by my co-founder Sam, just launched Working with Consultants, a free library of templates, guides, and case studies for nonprofits on the client side. Scoping, contracts, what to do when things go sideways. Worth a skim whether you're about to hire someone or already mid-engagement.
-Tareq