A nonprofit receives recurring questions about its campaign: When will the project begin? How can people donate? Where will implementation updates be published? The team knows most of the answers, but reviewing information and drafting responses takes time, especially during busy periods.
These tasks are among those now targeted by AI fundraising tools. In the education sector, technology platforms offer tools for drafting messages, summarizing interactions, and recommending which donors to contact, alongside agents that conduct outreach under institutional control. However, a provider’s description explains what its product offers; it does not establish that another nonprofit will save time or increase donations by using it.
The opportunity for nonprofits is to test one of these capabilities in their daily work. To produce a useful decision, the pilot needs a specific problem, reliable information, and a way to compare results with the team’s current performance.
Start With the Task That Needs Improvement
Suppose a nonprofit runs a campaign to fund an educational program, with one staff member answering donor inquiries. Some questions are straightforward, while others require input from the project manager or finance team.
Before selecting a tool, the organization reviews a sample of these inquiries: Which questions recur most often? Where is time being lost? Is it spent finding information, drafting responses, or waiting for an answer to be approved?
This assessment matters. If campaign information is incomplete or scattered across conflicting files, introducing another tool will not solve the underlying problem. The organization first brings approved answers together in a clear reference, then tests whether the tool adds value to the improved workflow.
The pilot’s objective can then be defined:
Reduce the time needed to prepare and review responses to recurring inquiries while maintaining accuracy and clarity.
With this objective, the organization knows what it is testing. It is not asking the tool to manage fundraising as a whole or assuming that faster responses will automatically generate more donations.
Define the Tool’s Role Before Granting Permissions
A drafting tool differs from an agent that performs steps within a workflow. The former suggests text for a staff member to use. The latter may be able to consult campaign information, prepare and send an answer, and record the interaction, depending on its capabilities and permissions.
The proposed pilot therefore begins with a limited role: the tool reads the question and prepares a draft using the approved reference. A staff member then reviews and sends the response. The nonprofit remains responsible for donor communication.
This arrangement allows the team to understand the tool’s performance before allowing it to send messages. It may draft a good answer about how to donate but make an error when asked about an unconfirmed date or a change to the purpose of a donation.
Cases requiring human handling are identified from the outset, including complaints, refund requests, and questions that could create a new commitment for the organization. If the reference does not contain an answer, the tool’s task is to refer the question to a staff member.
Give It Information the Organization Can Rely On
For the educational campaign, the reference includes the project’s purpose, intended beneficiaries, approved cost, current stage, confirmed dates, donation channels, and how donors can follow updates. A designated person updates it whenever information changes.
If a donor asks, “Will the program start next month?” but the start date has not been approved, the draft must not promise implementation that month. An appropriate response explains that the date has not yet been confirmed or refers the question to the team for verification.
This is where review becomes essential: generative AI systems can produce incorrect information in convincing language. The US National Institute of Standards and Technology identifies this issue, alongside data privacy and human oversight, among the risks that require management when using these systems.
Because the task is limited, its data requirements can also be limited. Answering a question about a project’s start date does not require the donor’s giving history or beneficiaries’ personal information. The nonprofit begins with de-identified questions and checks the tool’s terms for data storage, use, and deletion before introducing real personal data.
Test It on the Work Staff Already Perform
The nonprofit starts with a set of previous questions, including routine inquiries and cases without confirmed answers. A staff member reviews the drafts to check that the tool uses accurate information and refers cases beyond its scope.
If the results are satisfactory, the organization moves to a limited live pilot—for example, preparing inquiry responses over four weeks, with every draft reviewed before sending.
During this period, it compares its usual response-writing process with the AI-assisted process. Both use the same approved reference, and the questions are comparable in difficulty. This prevents the benefits of organizing information from being attributed entirely to AI.
The organization measures more than text-generation time. Timing begins when the question is opened and ends when the response is ready to send, including review and correction. A tool may generate a draft quickly but require extensive editing.
Alongside time, the organization records each draft’s quality: Is the information correct? Does it answer the question? Is the language clear and appropriate? Does it need minor editing or substantial correction? Tracking additional questions caused by unclear responses is also useful, as these can reveal problems that a text review alone might miss.
Assess Time, Quality, and Cost Together
Suppose, in a hypothetical educational example, preparing and reviewing 100 responses through the usual process takes ten hours. Completing a comparable batch with the tool takes six hours. The operational saving for that batch is four hours, provided response quality remains acceptable.
However, preparing the reference and configuring the tool took an additional five hours. The first pilot therefore required eleven hours in total, so it has not yet produced an overall time saving.
If subsequent batches continue to take only six hours, the benefit may begin to emerge. If the tool requires frequent updates and corrections, the saving may be smaller than it first appeared.
The nonprofit also reviews subscription and operating costs and how staff use the time made available. Removing four hours of work does not necessarily reduce payroll expenses, but it may allow a staff member to follow up on cases requiring personal attention or prepare better donor updates.
The result is valuable when the task becomes less burdensome at a reasonable cost while communication quality is maintained. Speed accompanied by errors or unauthorized promises does not meet the pilot’s objective.
When Should You Measure Its Effect on Donations?
A pilot focused on answering inquiries primarily measures service efficiency and quality. If the nonprofit wants to understand the tool’s effect on fundraising, it needs a separate test involving a clear donation request.
For example, it can compare follow-up messages prepared by staff with messages prepared using AI assistance, reviewing all messages before sending. For the comparison to be useful, the recipient groups should be comparable, while the campaign objective, sending time, and donation terms remain as similar as possible.
The nonprofit then measures completed donations and funds collected, alongside communication costs, unsubscribes, and complaints. Email opens and link clicks help explain engagement, but they do not equal completed donations.
If participant numbers are small or the groups face different conditions, the result should be treated as an initial indication. Likewise, a successful message for a public campaign does not establish that the same approach is suitable for a major individual donor or an institutional funder. Each relationship has its own context and commitments.
Let the Pilot Lead to a Specific Decision
At the end of the test, the nonprofit has three practical options: continue the limited use if there is a clear benefit, revise the reference and workflow before testing again, or stop if costs and corrections outweigh the value.
The organization may find that the tool is useful for preparing drafts while sending and follow-up still require a staff member. That is a complete decision grounded in what the pilot demonstrated. Granting the agent broader permissions requires a test proportionate to their consequences.
The nonprofit begins to benefit from this development when it can identify the work it wants to improve and verify that the tool has actually improved it. Through a small, measurable task, it can make a realistic decision about AI that respects staff time and preserves donor trust.
