a museum asked what AI could do for its fundraising. two calls, two decks, and a working prototype later, the answer was: one small job first, chosen by them, before any training.
WITH JASON MARCOTTE · PROPOSAL STAGE · NAMES REMOVED
grants brought in about $115,000 a year. finding funders they hadn’t tried was the first job they named for AI.
applications written fresh each time, tracked in spreadsheets.
event sponsors gave about $30,000 a year, mostly for a logo. they wanted partners all year.
a simple report from their database took one to two weeks.
What we agreed: the year’s budget was set, so start small, with one win on a real task before any training, and little of the team’s time.
about three weeks. before our second call, from public tax filings and funder websites alone, we found nine foundations that fit, and ruled out eight with the reason for each: one had stopped giving in 2018, two only fund other counties, two were already funders.
about two weeks. a full partnership case for one local business: a children’s dentist a quarter mile away that had just started taking the state’s low-income insurance. a year-round plan, four packages, and one check first: the museum already had a dental partner.
three routes compared, from training on a tool they liked to a custom bank of approved answers. they’d found ready-made grant tools lost their voice. a grant bank doesn’t write; it hands back what the team already approved.
each starting point stands on its own, with our hours and theirs named on one page. never a share of what they raise. a nonprofit rate, in exchange for this write-up.
most of the value came before any AI was switched on: hearing what was actually stuck, and saying back the numbers so they could correct them. the AI did the reading. a person checked every name, date and figure.