What AI Should and Should Not Do With Your Donor File

Where does AI genuinely help stewardship?

Drafting is the first job. First versions of thank-you letters, gift acknowledgments, appeal copy, and update emails are work AI does quickly and a person can edit in minutes. The draft is never the finished piece. A human edits it, makes it true, and signs it. The hours saved go back into the stewardship no tool can do, starting with the phone: the economics of a five-minute call are in our guide to thank-you call programs.

Pattern-finding is the second. The sector retains 43.3% of its donors year over year and just 18.9% of first-time donors, according to the Fundraising Effectiveness Project, and a file that leaks that fast rewards early flags: the monthly donor whose payment failed, the five-year donor who skipped a usual giving month, the first-time donor with no second gift nine months on. AI is well suited to surfacing those patterns from giving records. Choosing what to do about each one is judgment, and the taxonomy of why donors stop giving is the map for those choices.

The third job is preparation: summarizing a donor's giving history and past conversations before a call or a visit, so the person making the contact walks in informed. Every summary gets checked against the record before it is trusted, which is the subject of the next section.

Where should judgment stay human?

Start with verification. Vicki Pugh, CFRE, faculty at The Fund Raising School, put the essential rule plainly on a 2026 episode of the First Day Podcast: verify the results. AI tools hallucinate, and they present incomplete findings with confidence. Her discipline is concrete: compare outputs across tools, and keep trusted donor-research platforms and public records in the loop rather than treating a generated answer as settled fact.

Beyond verification, the moments that carry the relationship stay human. The upgrade ask, the legacy conversation, the call after a donor's spouse dies: these are exactly the moments donors remember, and they are the moments where a machine-written line does damage a discount never repays. A donor who learns a condolence note was generated does not remember the efficiency. The safe rule is the simple one: if the moment is why the donor stays, a person owns it.

What guardrails should a small organization write down?

One page is enough, and writing it puts an organization ahead of the 47% with no policy at all. Five lines cover most of the risk:

1. 1.Name what AI may touch and what it may not. Drafts and pattern-finding are in. Donor-facing sends and judgments about people are out.
2. 2.Keep donor records out of general-purpose tools. Names, addresses, and giving histories go only into systems the organization has vetted and holds a contract with.
3. 3.Assign a named reviewer. No AI-drafted text reaches a donor until a specific person has read, corrected, and signed it.
4. 4.Verify outputs against source records, and compare tools when the stakes are high, per Pugh's rule above.
5. 5.Write the workflow down. Ad hoc use, the mode 81% of the sector is in today, is how broad adoption produces thin results.

The point of the tools is time, and the point of the time is people. An organization that lets AI carry the drafts and the flags, keeps every judgment and every sensitive word human, and checks what the machine says against its own records will get the hours back and spend them where retention is actually won: on the phone, in the visit, in the thank-you that sounds like a person because it was one.