Until fairly recently, the scarce input in a fundraising meeting was ideas. Somebody had to think of something, and the quality of what a small organization did all year was constrained by the imagination of whoever was in the room.
That constraint is gone. Any generative model will produce a hundred fundraising ideas in a few seconds, ranked, formatted, and superficially plausible. This has been treated across the sector as a productivity gain, and it is one, but it has also moved the bottleneck somewhere most organizations are less equipped to handle. When generation is free, everything depends on selection.
This article covers why that shift happened, what the research says about how the sector is actually using these tools, and four tests worth applying to any idea before committing a quarter to it.
Generating ideas is no longer the constraint
The change here is not incremental. It is a change in kind, and it deserves stating precisely.
The supply of plausible ideas is now effectively unlimited
A language model has absorbed most of what has been published on fundraising, so it can reproduce the whole genre on request. Ask for twenty ideas for a rural health program and twenty arrive, each with a short rationale.
They are rarely wrong, and that is the difficulty. Drawn from the mean of everything published, they are generic in a specific way: they describe activities that worked somewhere, for someone, under conditions the model was never told about.
Which relocates the entire problem to selection
An organization can now hold forty viable-looking options and less basis for choosing between them than it had when it held two. Choosing badly is more expensive than it looks, because a fundraising idea consumes a quarter of staff attention and a portion of the goodwill an organization can only spend once.
Published idea lists were never designed to help you choose
Look at the standard article on this subject and it is a list, occasionally a long one, with no cost, effort, or suitability rating on any entry.
That format was already weak when ideas were scarce. With ideas now free, an unranked list adds very little. The more useful collections of fundraising ideas for public health causes are organized around work that nothing else funds, with each idea stating what the money buys.
More ideas than anyone can run
In many small organizations, AI use did not start with a policy or a training session. It started with one staff member trying a chatbot for a newsletter draft or a grant paragraph, then asking it for fundraising ideas because it was already open.
The ideas come back fluent and well organized, so they pass a quick read at a staff meeting. A list of twenty options looks like a plan, even when nobody has checked what each one would cost, who would run it, or whether the community has already been asked for the same thing.
Many fundraising plans now contain more ideas than the organization can execute, chosen for how good they sounded rather than against any stated criterion.
What is missing is a filter, and a model cannot supply one, because it is made of information about a particular organization in a particular place.
Four tests for choosing between ideas
Most ideas fail at least one of these quickly, and that saves time.
Does it fund a thing, or stage an occasion?
An occasion raises money once and consumes enormous effort: a dinner, a tournament, a gala. A thing is a defined piece of work that money buys and that continues afterward.
Occasions have their place, but they are usually chosen because they are familiar, not because the numbers favor them. Count the volunteer hours honestly before deciding.
Can you price a single unit of it?
If you cannot say what one unit costs, the idea is not ready. One month of a coordinator’s hours. One route run weekly for a year. Twelve appointments.
A unit price is the difference between a goal a donor can reason about and a number they have to take on faith. It also exposes ideas that sounded good and turn out to have no costable core.
Does something else already pay for this?
The strongest campaigns fund work with no other funding stream behind it. If insurance, a contract, or a formula grant already covers a service, donated money spent on it is subsidising an existing payer, and sophisticated donors notice.
Can your organization actually run it?
The last test is the one generative output cannot see. It requires knowing your volunteer base, your staff’s remaining capacity, your board’s appetite, and what your community has already been asked for twice this year.
An idea that fits a similar organization elsewhere may be entirely wrong here, and the idea itself will not show you that.
What judgment supplies that a model cannot
The distinction that matters is between knowledge that has been written down and knowledge that has not.
The local facts are not in any training data
Which employer in town gives, and to what. Which school principal will return a call. Which idea was tried four years ago and went badly enough that people still mention it.
That is the information selection runs on, it exists only inside an organization and its community, and no amount of model capability substitutes for it.
Constraints are the real input
A generative tool asked for ideas returns possibilities. The same tool given the actual constraints, the budget, the staffing, the calendar, the political sensitivities, becomes considerably more useful, because it is finally solving the problem the organization has.
Most disappointing output in this category comes from asking a broad question when the value was in the narrow one.
The part that stayed hard
Producing fundraising ideas was a real constraint, it has been substantially removed, and organizations that use these tools well save meaningful time.
What was never the constraint, and is now exposed as the binding one, is the decision. Choosing which single idea deserves the next quarter requires knowing a great deal about one organization in one place, and that has become more valuable rather than less as the supply of alternatives has expanded.
The organizations that will get the most from these tools are the ones that already knew how to say no to a good idea.
AUTHOR BIO
I’m Dr. Shital Sharma, founder of HealthCommons, a board-certified foot and ankle surgeon, and a healthcare leader focused on improving access to care. I earned my Doctor of Podiatric Medicine from Temple University and an Executive MBA and MS in Healthcare Leadership from Cornell University. I’m also pursuing an MPH at Yale School of Public Health. Here I write about the systems that shape access to care, the public health challenges communities face, and the local solutions that address them.