AI for program evaluation
Logic models, outcome frameworks, and making sense of data you're already collecting.
Evaluation is the function where nonprofits most often have the raw material and least often have the capacity to use it. Survey responses pile up, attendance gets tracked, and nobody has three uninterrupted days to work out what any of it means.
This is a genuinely good fit for AI assistance, with one significant caveat about which parts of the analysis you can trust.
Where it helps most
Logic models from a program description.
Describe what the program does and ask for inputs, activities, outputs, outcomes, and impact. You'll edit it, but starting from a structured draft beats starting from the framework template.
Theory of change articulation.
Most organizations have one implicitly and struggle to write it down. Talking it through and asking for a written version is faster than a retreat.
Coding open-ended survey responses.
Paste anonymized free-text answers and ask for the recurring themes. This is legitimate qualitative work and it's the single biggest time save in the function.
Survey design.
Ask it to review your draft questions for leading language, double-barrelled items, and response scales that won't tell you anything.
Translating evaluation findings for different audiences.
The same results written for a funder report, a board summary, and a community newsletter.
Where to be careful
Statistical analysis.
These tools will produce confident-sounding statistical claims that don't hold. If a finding is going into a funder report, the analysis needs a person who understands the method.
Attribution claims.
AI will happily write that your program caused an outcome. Whether it did is a research design question, and overclaiming in an evaluation is the kind of thing that damages credibility for years.
Client data.
Anonymize before anything goes into a general tool, and check that free-text responses don't contain identifying detail — people name their caseworkers and their neighborhoods in open comment fields more often than you'd expect.
Add a real example here — ideally survey analysis or a logic model build with a real organization.
The Organizer handles documentation and framework work, including logic models and SOPs. The Scorekeeper picks up anything involving numbers and trends.