Donor Consent and Data Ethics in Machine Learning Fundraising
Machine learning has made its way into fundraising quietly. Many CRM platforms now include features that predict who is likely to give, suggest ask amounts, estimate wealth or flag donors at risk of lapsing. These tools can help a stretched development team focus its time. They also run on personal data that donors may never have expected to be used this way.
Donors give because they trust an organization. That trust extends to how the organization treats their information: their address, their giving history, their interests, sometimes their family details. When that data feeds a model that profiles and ranks them, the question is not only whether it is legal. It is whether donors would be comfortable if they knew.
This article sets out a practical approach to consent and data ethics for nonprofits using machine learning in fundraising, from what to collect to how to explain it.
How Machine Learning Uses Donor Data
It helps to be specific about what these systems actually do, because “AI” covers very different practices.
| Use | Data typically involved | Main ethical concern |
|---|---|---|
| Propensity scoring | Gift history, engagement, event attendance | Donors ranked without knowing it |
| Wealth screening | Public records, property, third-party data | Data combined from sources donors never shared |
| Lapse prediction | Giving frequency, email activity | Tracking behavior donors may consider private |
| Personalized messaging | Interests, past responses, demographics | Manipulative targeting or sensitive inferences |
Inferences are data too
A model may conclude that a donor is likely elderly, recently bereaved or financially stressed, even if the donor never said so. These inferred traits can be more sensitive than anything the donor shared directly. Ethical practice treats them with the same care as personal data, and often more. The tension between relevance and intrusion is explored further in the promise and peril of personalization in AI-powered fundraising.
What Meaningful Consent Looks Like
Consent in fundraising is often reduced to a checkbox on a donation form. For machine learning, that is rarely enough.
Legal minimums versus ethical standards
Rules differ by country and state. In Europe, GDPR requires a lawful basis for processing and gives people rights to object to profiling. In the US, privacy laws vary by state and are changing. Many nonprofits rely on “legitimate interest” or broad privacy notices. Meeting the legal minimum is necessary, but donor trust usually depends on a higher standard: would a reasonable donor be surprised by what you are doing?
Clear, specific and reversible
Meaningful consent has a few consistent features:
- Specific: it names the purpose, such as “to tailor which appeals you receive,” rather than a vague “to improve our services.”
- Understandable: it uses plain language, not a reference to a long policy.
- Separate: agreeing to give is not bundled with agreeing to be profiled.
- Reversible: donors can opt out of profiling later without stopping their support.
Data Minimization and Purpose Limits
The simplest way to reduce ethical risk is to use less data. Machine learning tempts teams to feed in everything available on the theory that more data means better predictions. Often it means more risk with little gain.
Ask what each field is for
Before a dataset goes into a model, review each field and ask whether it is needed for the stated purpose. Health information, religious affiliation, and details about family members rarely belong in a fundraising model. Neither do notes written by staff for relationship purposes, which may include personal remarks never meant for analysis.
Be cautious with third-party data
Wealth screening and data appending combine your records with outside sources. Donors who shared an email address to receive a newsletter did not necessarily expect their property value to be looked up. If you use these services, disclose it, limit it to the purposes you have explained and check the vendor’s own data sources.
Set retention limits
Model outputs, such as scores and segments, should not live forever. Decide how long scores are kept, when they are refreshed and when inactive donors’ data is deleted.
A good test for any donor data practice: could you describe it in your annual report without donors feeling uneasy?
Governance: Who Decides and Who Checks
Data ethics fails most often not because of bad intentions but because nobody owns the question. A development team adopts a CRM feature, a vendor turns on a scoring model, and no one reviews whether it fits the organization’s values.
- Assign a named person responsible for donor data ethics, even in a small team.
- Keep a simple inventory of which models and features are active and what data they use.
- Review new AI features before they are switched on, including those added by vendor updates.
- Check periodically whether scoring treats groups of donors unevenly, such as small recurring givers.
- Report to the board at least once a year on how donor data is used.
These steps fit naturally into a wider organizational policy. If you are drafting one, a practical framework for building a nonprofit AI policy covers the broader structure.
Questions to ask your CRM or analytics vendor
Ask whether your donor data is used to train models for other clients, where it is stored, whether you can turn off individual AI features, and how you can export or delete data. Get the answers in writing.
Explaining Your Practices to Donors
Transparency is the part of data ethics donors actually see. It does not need to be elaborate.
Write a short data note
Add a brief section to your privacy page, written in everyday language, that explains whether you use predictive tools, what they are for and how to opt out. Imagine a regional arts organization adding three sentences: it uses giving history to decide which appeals to send, it does not buy outside data about donors, and anyone can ask to be excluded. That is clear and honest, and it takes minutes to read.
Make opting out easy
Offer a simple way to opt out by email or through a preference center, and honor it fully, including in any vendor systems.
Consent as Part of the Donor Relationship
It is easy to see consent and data ethics as compliance tasks that slow fundraising down. They are better understood as part of stewardship. Donors who feel respected are more likely to stay, to give more over time and to recommend the organization to others. Donors who discover they have been profiled without their knowledge often feel the opposite.
Machine learning can help development teams spend their time well and reach supporters with messages that matter to them. It works best when donors know what is happening, have a real choice and see that the organization uses their information with the same care it brings to its mission. That is the standard worth holding: not just what the data allows, but what the relationship deserves.
