All case studies

Build and Run · Nonprofit

How a regional education nonprofit got its funder research capacity back with AI

At a glance

  • About 3 minutes for a full funder profile, in the format the fundraising team already uses.
  • 61 of 105 prospect organizations profiled by October 7. A 43-company list and 26 school districts were researched in bulk, all complete.
  • 5,584 past gifts matched onto 60 prospects. Latest fiscal-year totals for 44 of the top 47 funders match the team's own report to the dollar.
  • 6 of 9 profiles in an early-September benchmark found contact emails equal to or better than the hand-written version.

Who they are

A regional youth education nonprofit in the United States. Its advancement team raises money from companies and foundations. A technology partner that advises the nonprofit sold and coordinates the work, with Vixure building and running it.

What they needed

The nonprofit once had a small team that did all prospect research. That team was absorbed into other roles. At the kickoff, the advancement lead explained why they wanted a tool: "we've lost this capacity on our team." The goal: every fundraiser researching funders on their own.

Good research meant reading five years of a foundation's tax filings, finding the right door into a company, and checking internal rules before asking. In the advancement lead's words: "I know most of my team members do not spend a bunch of time in 990s."

What we built

Account intelligence for funders, with a second engine for schools.

Funder research. Eight AI agents each own one part of the profile, following the method the advancement team described, from how to apply to five years of grants and the right contacts. Two kinds of facts are read by code, because models kept missing them: grant tables in tax filings and contact details published on the funder's own site. The agent that writes the verdict has no web access, so it reasons only over facts other agents found and cited. A quality agent names the weakest section, and that part is researched again.

Each profile shows a verdict (worth pursuing, worth a look, or not a fit right now), a 12-signal checklist, and a flag when an approach needs internal sign-off first. Gift and volunteer history from the nonprofit's own systems is matched onto the profiles, with personal data dropped before loading.

Outreach. AI drafts an outline for a first message: the angle, the fact to open on and the ask. A person at Vixure reviews it. A fundraiser turns it into a message, written by hand or with AI, and only a signed-in person can approve it. Nothing goes out without a person.

Schools. A second engine holds districts, schools, enrollment, seven years of program history and contacts.

How we run it together

A Vixure project lead runs two-week sprints that close with a client review. The partner is the single point of contact and sets priorities. Twelve people have accounts on the live platform.

Improvements since the first release on September 3:

  • research runs that finish in about 3 minutes in production (September 13)
  • a reviewer's feedback on nine test profiles, built the same day, including a verdict weighted by who a funder already gives to (September 21)
  • team labels and owners on each prospect, and a first draft of the schools engine (October 1)
  • gift and volunteer history matched onto the profiles (October 7)

The value

Before: research lived with a small team that has since moved to other roles, and each profile was written by hand. Quality is measured against nine of those hand-written profiles.

After: any fundraiser can open a full profile, built in about 3 minutes. In the early-September benchmark, the engine found every listed grant on four of the nine. On one family foundation it returned four staff email addresses, five application windows and 643 grants, where the hand-written profile had one address, one window and 36 grants.

What's next

The schools team reviews the second engine from October 12 to 16. The release due October 23 adds a contact log and access set per person. Deeper research on the top funders follows.

Client details are anonymized. Figures come from project records as of October 11, 2026; estimates are marked.

What's the most expensive workflow you're still running by hand?

Book 30 minutes and we'll show you what we would build.

Book a Call