What AI can free up and what it should not touch

A Tuesday morning, and nobody decided

Picture a development office on an ordinary Tuesday. The annual fund coordinator opens a draft appeal that a tool wrote overnight from last year's letter and this year's numbers. The gift officer pulls up a prospect summary that a system assembled from wealth screening, past giving, and board notes. The stewardship assistant sends a thank-you letter shaped by a template that adjusts its own sentences to the size of the gift. Nobody in that office held a meeting to approve any of this. It arrived a plug-in at a time, a renewal at a time, and now it is simply how the work gets done.

That is the starting condition for this module, and it is worth saying plainly: the question in front of you is not whether your shop will use AI. In some form, most already do, and the rest will soon enough. The real questions are narrower and harder to duck. Does the donor stay free to say no, or not yet, without ever knowing a machine had a hand in the ask? Does the letter that reaches them say something a person actually means? And when the piece goes out with an organization's name on it, whose name is on the work behind it?

Notice what did not happen in that Tuesday morning scene. No one sat down and asked whether a machine-drafted appeal changes what the donor is owed under the Code. No one asked whether a system-built prospect summary crossed the line between knowing a person and processing one. Those questions did not get skipped because the office is careless. They got skipped because each individual tool arrived looking small: a faster draft, a tidier summary, a template that saves twenty minutes on a Friday. None of that felt like a decision. Add them up, though, and a shop can wake up running on judgment it never actually exercised.

This lesson gives you a way to answer those questions that does not depend on knowing how any particular tool works, or which one your database vendor bundles in next quarter's update. It depends on the Framework you already have. The seven movements were built to protect one thing, and that one thing does not change because the tool changed.

The test: run it through the seven movements

The Invitation Framework was built to protect one thing: a gift has to stay voluntary, which means the donor has to stay free through every movement, from Pause to Multiply. That gives you a test for AI that does not go stale when the tools change. Ask, for any use you are considering, which movement it touches, and whether it is helping a person do that movement or replacing the person who should be doing it.

Pause. This is where AI is most clearly welcome. Pause asks the fundraiser to arrive honest: what am I carrying into this, do I have the facts right, am I free to hear no. A tool that summarizes a donor's giving history so the gift officer walks in prepared, or that flags a scheduling conflict before a visit, is serving Pause. It is getting the fundraiser ready. It is not doing the asking.

See. See asks you to know the person, not the profile. A tool may serve See when it organizes information the donor would expect you to have and already gave you: gift history, event attendance, a note from a prior conversation. It stops serving See the moment it infers things the donor never told you (assumed income, family situation, health) from behavioral exhaust the donor did not know you were collecting. Knowing the record is not knowing the person, and a tool that only touches the record has to be labeled as exactly that.

Discern. Discern tests whether an invitation is true and ready: is this the right ask, for this person, right now. AI can help you rehearse the conversation, draft two or three versions of an ask at different levels to think through, or check a case statement against the organization's own numbers before it goes to print. It cannot tell you whether a relationship is ready, and it has no way to weigh what the gift officer heard in the donor's voice at the last visit, or the fact that the donor's spouse just lost a job, or the dozen small signals a person picks up and a system never will. That judgment stays with the fundraiser who has sat across from the donor.

Invite, Receive, Accompany. These three are where the line holds firmest, because each one is the relationship itself, not preparation for it. Invite names the mission connection and leaves the donor free; that has to come from a person who means it. A draft can suggest language, but the person who delivers the ask has to be able to answer the donor's next question, in the moment, without reading from a screen someone else filled in. Receive honors whatever answer comes, including no and not yet, in real time, with the judgment a moment like that requires; a system does not hear the hesitation in a donor's voice or know when silence means think it over rather than no. Accompany keeps the promise made at the ask, over months or years, which means a human has to know what was promised and check, personally, that the mission held up its end. A tool can draft language for any of these three movements. It cannot be the one standing in the relationship, and the moment a donor cannot tell whether they are talking to a person or a system that is imitating one, something has already gone wrong, whatever the letter says about impact.

Multiply. Multiply builds other people's capacity: training a board member to make an ask, coaching a new officer through their first difficult conversation. AI can help you build training materials, draft practice scripts, or organize what a new hire needs to read in the first month. It does not get to decide who a person is, what they are capable of, or what they are ready to be invited into. That decision, like Discern's, is a human one, made by someone who has watched the person work.

The pattern across all seven movements is the same. AI may prepare, organize, and rehearse. It may not decide who someone is, and it may not stand in for the relationship. Keep that sentence in view and you can evaluate a tool you have never seen before, the day it shows up in a vendor's release notes, without waiting for a committee to study it first.

Four ways AI use fails a donor

Most failures in this area are not dramatic. Nobody sets out to deceive a donor. They are small drifts that each seemed reasonable on the day, made by tired people trying to get a Friday deadline met, and they cluster into four patterns worth naming so you recognize them before they become the way your office works.

Synthetic sincerity. A thank-you letter goes out that performs gratitude nobody actually felt: warm language, a personal-sounding detail, a closing line that reads like it came from someone who was moved. No one was moved. No one read it before it sent. The donor cannot tell the difference from the page, but the relationship the letter claims does not exist yet, and eventually that gap shows, usually at the worst possible moment: a follow-up call where the donor references a detail from "their" letter and the officer on the phone has no idea what they mean. The fix is not a ban on drafting tools. It is a rule: a person means what is sent under their name, and reads it before it goes, and the shop says so when a tool helped shape a first draft. Meaning the words and disclosing that a tool helped are not the same requirement. You need both, and neither one substitutes for the other.

Reduction of the person to a score. A predictive model rates a donor's "propensity" or estimates their capacity, and somewhere along the way the score stops being a hypothesis and starts being treated as a verdict: an officer skips the relationship-building because the score is low, or over-asks because the score is high, or a stewardship plan gets built entirely around a number instead of a conversation. The worksheet on analytics policy in the Philanthropy Analytics module works through this in detail and this lesson will not repeat it; the short version that belongs here is that a score opens a question a human asks. It never closes one. And some questions, about health, family circumstance, or anything the donor did not tell you and did not make public, are not asked at all, by a model or a person, no matter how confidently a vendor's dashboard offers the answer.

Donor data leaving your custody. A gift officer, trying to write a better solicitation letter under deadline, pastes a donor's full giving history, private notes, and contact information into a general-purpose tool that was never vetted for that purpose, to get a faster draft. The data has now left the organization's custody, on terms nobody at the organization reviewed, held by a party who made no promise about what happens to it next. This is not a hypothetical; it is one of the fastest-growing ways a development office breaches its own privacy notice without anyone intending to, one paste at a time. The fix is custody rules written down before the first draft, not discovered after an incident: which tools are approved for donor data, what vendor terms were actually read by someone qualified to read them, and what never leaves the system of record no matter how much faster the workaround feels on a Tuesday afternoon.

Accountability with no one's name on it. A piece goes out, something in it is wrong or overstated or, worse, simply untrue, and when someone asks who is responsible, the honest answer is that no one really checked. The tool produced it, the queue moved it along, and each person who touched it assumed the step before theirs had caught anything wrong. That assumption, repeated down a chain of hands, is where accountability actually breaks. The fix is the simplest rule in this lesson: every AI-assisted piece has an owner, a named person who read the final version and would defend every sentence in it to the donor's face, in person, without a disclaimer. If no one can name that person, the piece does not go out.

What the donor is told, and when

The Donor Bill of Rights gives donors the right to know who is asking and to expect prompt, truthful answers to their questions. The AFP Code asks for accuracy in what you tell donors and confidentiality in how you handle what they give you. Neither document was written with AI in mind, and neither one needs to be rewritten because of it. Put together, they already set a disclosure standard: a donor who asks should get a plain answer, and your shop should not need to improvise one under pressure, in the middle of a call, with a donor waiting.

Here is a sentence a shop can actually use, at the point a donor asks or a policy requires disclosure: "We use tools to help us draft communications and organize donor information faster, and every piece you receive from us is reviewed and approved by a member of our team before it's sent." That sentence is true only if your practice matches it, word for word, every time. It does not promise no AI was involved anywhere in the process. It promises what the donor actually needs to know: a person stands behind what reached them, and nothing about their information was handled in a way the donor was not told about.

Say that sentence out loud in your own office before you rely on it. Ask the coordinator who drafts the appeals whether it is true. Ask the officer who pulls the prospect summaries. Ask whoever answers the phone when a donor calls with a question about a letter they received; that person needs the sentence too, ready, not improvised under pressure. If it is not true today, that is the gap to close, not the sentence to soften, and closing it is exactly the work the next three lessons walk through.

Why this shop leads, and how

There are two easy paths here, and neither one is the standard. One easy path is refusing the tools outright, which trades a real risk for a real cost: slower drafts, thinner prospect research, donors who get less attentive stewardship because a person is doing by hand what a tool could safely have prepared in a fraction of the time. A shop that refuses on principle still has to explain, to its own board, why donors are waited on more slowly than they need to be. The other easy path is adopting whatever the field adopts, at the pace the field adopts it, because everyone else is, which is exactly the pressure that produced the Tuesday morning at the top of this lesson: tools arriving one at a time, each one reasonable in isolation, until nobody remembers deciding and nobody can say why any particular use is in place.

The standard this shop holds is neither. It is a standard you can read aloud to a donor, in the donor's own language, without wincing and without a lawyer standing behind you: here is what we use these tools for, here is what we never let them do, and here is who is accountable for what you receive from us. That standard does not require you to be first with every new capability, and it does not require you to sit out while the field moves. It requires you to be able to say, plainly and specifically, what your shop decided and why. A shop that can say that sentence and mean it, in front of a donor, a board member, or a reporter, has already done most of the work this module asks of you. The rest is writing it down and keeping it current.

Three lessons follow this one. The next maps where AI belongs, and does not, inside each of the seven movements, in the kind of practical detail a gift officer can use on a Tuesday. After that comes the written policy itself: the document your shop adopts, keeps on file, and can hand to a donor, a board member, or an auditor who asks. And the module closes with an audit: naming every AI use already running in your office today, rating what it touches and what decision it shapes, and deciding, tool by tool, what to keep, what to change, and what to stop.

Evidence and adaptation note

This is a working tool, not a universal benchmark. Replace every bracketed field and example number with your organization's facts. Composite cases are labeled; their figures illustrate the method and should not be cited as sector results. Check legal, tax, privacy, employment, and accounting language against current guidance and your jurisdiction before adoption.

Primary references for review

Use these as verification starting points. The named reviewer still owns the final interpretation and must confirm that each source is current.

• Association of Fundraising Professionals, Code of Ethical Standards: https://afpglobal.org/ethics/code-ethical-standards
• Association of Fundraising Professionals, A Donor Bill of Rights: https://afpglobal.org/donor-bill-rights
• National Institute of Standards and Technology, AI Risk Management Framework: https://www.nist.gov/itl/ai-risk-management-framework
• Fundraising.AI, responsible AI framework: https://fundraising.ai/