The Questions Nonprofits Stopped Asking About Their Data

TL;DR: Nonprofits used to ask sharp, uncomfortable questions about the data they collected — who it served, who it left out, and why we were gathering it at all. As the tools got slicker, those questions got quieter. This is a plea to ask them again, before AI turns our unexamined data into unexamined decisions.

I have been in a lot of rooms where a nonprofit proudly shows me its dashboard. The charts are clean. The numbers move in the right direction. And when I ask why a particular field is being collected — what decision it informs, who benefits when we know it — the room goes quiet. Not because the answer is secret. Because nobody has asked in a while.

That silence is what worries me. Not the data itself, and not the dashboards. The fact that we stopped interrogating our own collection habits and started treating them as furniture — just there, part of the room, not something you question. And now we are pointing AI at all of it, which means the questions we skipped are about to get answered for us, at scale.

Why did nonprofits stop asking about their data?

I don’t think it was carelessness. I think it was momentum. Fifteen years ago, collecting data felt like a deliberate act — you built a form, you justified every field to a skeptical program director, you knew exactly why the ethnicity question was there and what it would and wouldn’t be used for. Collection was expensive, so it was intentional.

Then the tools got cheap and generous. Your CRM offered a hundred fields out of the box. Your event platform captured things you never asked for. Integrations quietly piped data from one system to another. Collection stopped being a decision and became a default. And defaults don’t invite questions — that is what makes them defaults. So the reflective muscle atrophied: the habit of pausing before a new field and asking whether it earns its place.

What were the questions we used to ask?

They were plain questions, and they still work. Why are we collecting this? What decision does it change? Who is served when we hold this information — and who is exposed? What happens to someone we never counted because they didn’t fit our form? Would the person this data describes recognize themselves in how we use it?

Notice that none of these are technical. You don’t need a data scientist to ask them — you need someone willing to sit with the discomfort of not having a good answer. That, more than any tool, is what has gone missing. In my AI Equity Framework I call the underlying layer Data Equity, but the framework is just scaffolding for a practice that used to be ordinary.

But isn’t more data just safer to keep?

This is the honest counterargument, and I want to give it its due. More data feels like insurance. You can’t analyze what you didn’t collect, so collecting broadly seems like protecting your future options. And there is a real cost to under-collecting — communities have been harmed by being invisible in the data as much as by being over-surveilled.

I hold that tension. I don’t think the answer is to collect less as a reflex, any more than to collect more as a reflex. The answer is to make it a decision again — to be able to say, out loud, why each thing is in your system. “We might need it someday” is not a reason. It is the absence of one. And every field you hold that you cannot justify is one that will shape an AI model’s output without ever passing through your judgment.

What happens when AI reads the data we stopped questioning?

Here is the part that keeps me up. A dashboard shows you what you put into it, and a person still reads it. AI is different — it finds patterns in everything: the fields you forgot were there, the proxies you didn’t know were proxies, the historical bias baked into who you served and how you recorded it. It doesn’t skip the unexamined data. It weights it.

So the ZIP code you collected for mailing labels becomes a stand-in for race. The “engagement score” nobody defined becomes a filter for who gets outreach. The gaps in your data become gaps in what the model believes is possible. None of this is the AI being malicious. It is the AI being faithful to data we never audited. Garbage in, gospel out.

How do we start asking again?

You don’t need a task force or a new platform. You need to make a few small commitments and keep them. Here is where I’d start:

• Audit one system, not everything. Pick a single database and walk field by field. For each one, name the decision it informs. The fields with no answer are your first conversation, not your last.

• Write down the ‘why’ next to the ‘what.’ Every field you keep should carry a one-line reason a community member would find reasonable. If you can’t write the line, you’ve found something to reconsider.

• Ask who is missing before you ask what’s trending. Absence shapes every analysis. Make “who isn’t in this data” a standing question in any meeting where the data drives a decision.

• Invite the people the data describes. Co-design isn’t a workshop you run once. Show the communities in your data how you use it, and let their discomfort revise your practice. That is Power and Self-Determination, made concrete.

• Set a retention line and honor it. Data you no longer need isn’t an asset — it’s a liability waiting for an algorithm. Decide what to delete, and actually delete it.

I don’t have this fully figured out, and I distrust anyone who says they do. But I am sure of one thing: the questions we stopped asking didn’t become less important — they just got louder while we weren’t listening.

---

Frequently asked questions

What does “data equity” actually mean for a small nonprofit?

Data equity means the way you collect, hold, and use data reflects fairness to the people that data is about — not just efficiency for your organization. For a small nonprofit, it is less about advanced technology and more about intention: knowing why each field exists, being honest about who your data represents and who it misses, and treating the communities in your records as participants in the analysis rather than subjects of it. It is a practice you can begin with the systems you already have.

Should we collect demographic or identity data at all?

Often yes — not collecting it can render communities invisible and make inequities impossible to see or address. The problem is rarely collection itself; it is collecting without a clear purpose, without consent, or without a plan for how the data protects rather than exposes people. The better question is not whether to collect identity data but whether you can state plainly why you collect each piece, how it will be used, and how the people it describes had a say. Answer that, and collection becomes an act of respect rather than surveillance.

How does unexamined data affect an AI tool we might adopt?

AI systems learn from the data you give them, including the fields and biases you may have stopped noticing. If your records over-represent some communities and under-represent others, or contain proxies like ZIP code that correlate with race, an AI tool will treat those patterns as truth and reproduce them — often faster and more confidently than a person would. Auditing your data before adopting AI is not a technical nicety; it is the difference between a tool that supports your mission and one that quietly scales your blind spots.