Start with Impact, Not Tech: A Framework for Evaluating AI Use Cases at Your Nonprofit
The most common mistake nonprofits make when exploring AI isn’t choosing the wrong tool. It’s choosing a tool before they’ve decided what they’re actually trying to accomplish.
“Start with impact, not tech” sounds obvious. In practice, most AI conversations at nonprofits start with tools — someone heard about ChatGPT, a board member read about AI, a funder mentioned it in a briefing — and the organization finds itself evaluating technology before it’s asked the more fundamental question: what problem are we actually trying to solve, and is this the right kind of solution?
This article gives you a framework for answering that question rigorously. It’s not complicated. But it changes what you decide to build, and what you decide to skip entirely.
The Three-Category Framework
Not all nonprofit work is equally appropriate for AI assistance. Trying to treat it as a single category — either “AI can help with everything” or “AI is too risky for our work” — misses the actual shape of the opportunity.
A more useful frame divides work into three categories:
Category 1: Automate
Tasks that are routine, high-volume, low-stakes, and well-defined. These are tasks where the main cost is time and where errors are caught easily before they cause harm.
Examples in nonprofit contexts:
- Summarizing meeting notes and distributing action items
- Generating first drafts of standard communications (donor acknowledgements, volunteer updates, program newsletters)
- Categorizing and tagging incoming data (survey responses, intake forms, grant eligibility screening)
- Creating internal documentation and FAQ resources from existing materials
- Research synthesis — pulling key points from long reports, policy documents, or sector literature
AI handles these well because the tasks are well-defined, the outputs can be reviewed by humans before they go anywhere consequential, and the failure mode (a draft that needs editing) is recoverable.
Category 2: Augment
Tasks that require human judgment, expertise, or relationship — but where AI can meaningfully support or accelerate the human doing the work.
Examples in nonprofit contexts:
- Grant writing — AI drafts, human refines with organizational knowledge and relationship context
- Donor research — AI compiles and synthesizes publicly available information, fundraiser interprets and decides
- Program evaluation — AI helps analyze qualitative data patterns, program staff provide contextual interpretation
- Strategic planning — AI synthesizes research and generates options, leadership decides
- Staff training development — AI builds first-draft curriculum frameworks, subject matter experts shape content
The key distinction here is that the AI is assisting a human judgment call, not replacing it. The human remains the decision-maker. The AI’s contribution is to make that human faster, better-informed, or less exhausted — not to substitute for their expertise.
Category 3: Human-Only
Tasks where AI assistance would undermine the quality of the outcome, the trust of the people being served, or the values of the organization — regardless of technical capability.
Examples in nonprofit contexts:
- Direct client relationships, particularly with vulnerable populations
- Clinical or therapeutic work
- Hiring decisions about individual candidates
- Disciplinary or termination decisions
- Situations where the person involved has a reasonable expectation of private, human-to-human engagement
- Any decision with significant consequences for an individual that they haven’t consented to have AI involved in
This category will evolve as tools improve and as norms develop. But in 2026, the organizations getting this right are erring on the side of keeping direct human work human — using AI to free up capacity for that work, not to substitute for it.
The Mission Integrity Filter
Before moving any potential AI use case into your “Automate” or “Augment” categories, apply one additional test:
Would doing this with AI undermine the reason this work matters?
This question does something the efficiency lens doesn’t. It forces you to ask whether the way you do something is part of what makes it meaningful.
A handwritten thank-you note from a development director has a different relationship to the donor than a personalized AI-generated note, even if the content is identical. A program intake conversation handled by a trained staff member signals something different to a client than one processed by a chatbot. A peer support session facilitated by someone with lived experience carries a meaning that AI assistance might dilute rather than amplify.
None of this means the answer is always “don’t use AI.” It means the efficiency calculation isn’t the only calculation. Organizations that skip this filter tend to optimize their way into damaging the very things that made them effective in the first place.
The Diagnostic Questions
When evaluating a specific AI use case at your organization, work through these questions in order:
1. What specific problem are we trying to solve?
Be precise. “We want to use AI for communications” is not a problem statement. “Our communications coordinator spends 8 hours per week on first drafts of standard donor updates that don’t require significant customization” is a problem statement.
2. Is this actually a technology problem?
Some of what looks like an efficiency problem is actually a process problem, a capacity problem, or a clarity problem. AI won’t fix unclear roles, poor communication, or misaligned priorities — it sometimes makes those problems more visible and occasionally makes them worse. If the root issue isn’t the time spent on a task but something upstream of the task, solve the upstream problem first.
3. Which category does this work fall into?
Automate, Augment, or Human-Only. If you’re not sure, default to Human-Only until you have more information.
4. Does it pass the Mission Integrity Filter?
Would doing this with AI compromise the meaning or quality of the work in ways that matter to the people you serve?
5. What does success look like, and how will we measure it?
Before implementing anything, define what you’re trying to achieve and how you’ll know if it’s working. Time saved? Quality improved? Staff capacity freed for other work? Without a clear metric, you can’t evaluate whether the implementation was worth it.
6. Who needs to be involved?
Specifically: the staff who will use the AI tool, the clients or communities affected by the outputs, and whoever holds organizational accountability for the relevant decisions. Any AI implementation that doesn’t involve the people closest to the work tends to fail or get abandoned.
A Note on Vendor Claims
AI tool vendors will almost always frame their products as appropriate for your highest-value, most complex work. That framing serves their interests, not yours.
The organizations getting the most durable value from AI are typically using it for their most routine, lowest-stakes work first — and expanding carefully from there. This isn’t technophobia. It’s good implementation practice. Low-stakes pilots let you learn what works at your organization before the stakes are high.
“What can AI do?” is the vendor’s question. “What should we use AI for, given our specific situation?” is yours.
FAQ
How do we know if we’ve categorized something correctly?
The best test is to ask the staff who do the work and, where possible, the people who receive it. The people closest to the task have the clearest view of what the human element is actually doing and whether it’s replaceable in a given context.
What if we disagree internally about which category something belongs in?
That disagreement is productive. It usually surfaces assumptions about what the work is actually for and who it serves. Surface it explicitly rather than letting one faction make the call unilaterally.
Should we involve clients in these decisions?
Where practical, yes — particularly for Augment and Human-Only categories. For organizations serving vulnerable populations, involving clients or community members in conversations about how AI will or won’t be used is both a values question and a trust question.
How often should we revisit these category assignments?
At minimum annually. AI capabilities are changing quickly, and a task that belongs in the Human-Only category today may shift as tools improve and as your organization develops clearer norms and practices.
Mitch Schwartz is the founder of Ops Machine, a Montreal-based AI integration and workflow consultancy. He works with nonprofits and organizations mid-transformation to find where AI fits, build the right systems, and make sure teams actually use them. Book a free discovery call →