AI can help a nonprofit team review attendance, staff notes and participant feedback together, identify questions worth investigating, and prepare information for follow-up. Start with one recurring decision your team can act on. Connect the records needed to answer it, check AI's interpretation with staff and participants, and test a small change. The useful result is a better way of running the program, with evidence of what happened after the team acted.
Unmesh Sheth explains seven practices for small teams, drawing on Marco's account of Open Play Foundation. The session-timing experiment is a teaching example, not a reported Open Play result.
What should your team do first?
Use these seven practices in order:
- Choose one operational question and name the person who can act.
- Bring the field team into the review of what it collects.
- Read numbers alongside participants' explanations.
- Keep the relevant records and history connected.
- Test one manageable change and check what follows.
- Prepare different audience updates from the same reviewed evidence.
- Expand only after the first workflow is useful and maintainable.
1. Choose a question someone can act on
“Use AI” does not tell a program manager what to do on Monday. “Why are people missing the evening sessions, and should we change the timetable?” identifies a decision, a group and a possible response.
Write the question, decision owner and review date before collecting more information. Agree what would count as a useful answer and what would make the team reconsider its first explanation. A spreadsheet and a short review meeting may be enough for the first test.
2. Include the people who collect the information
Field staff should be able to examine what their records show, explain unfamiliar patterns and question an AI summary. A coach who spoke with a family may know that an attendance gap reflects a changed schedule rather than loss of interest.
Give staff an appropriate view of the records and a clear route to correct them. Ask an analyst to help with difficult comparisons or uncertain findings. Collection, review and action need named responsibilities, even when one person holds several roles.
3. Read attendance and explanations together
Attendance identifies a pattern. A conversation may help explain it. Keep each source's date, author and relevant program context so a current absence is not explained using an unrelated comment from months ago.
Where appropriate and permitted, connect records using a stable participant identifier. Keep a participant's own words distinct from a staff interpretation. Anonymous feedback can inform a group decision without identifying respondents.
In the video, an assistant suggests that late sessions might make girls' journeys home less safe. That is a question to investigate. Staff should check the supporting records and speak appropriately with affected people before drawing a conclusion or changing provision.
4. Test the complete workflow before choosing a system
Ask a prospective tool to handle one real job: capture an attendance record, connect a relevant note, help a reviewer investigate the pattern, record an action and retrieve the later follow-up.
Check whether the answer shows its supporting records, whether your staff can correct a mistake, and who maintains definitions and access. Include setup, integration, review, training and support in the cost. Keep specialist operational systems where they are needed; choose the arrangement that makes the full job workable for your team.
5. Run a small experiment
Illustrative example, not an Open Play result: staff suspect a session's timing is a barrier. After checking with participants, they test an earlier session at one site. They record who was expected, who attended, what families reported and other changes that could affect attendance.
The team agrees a review date and a decision: keep, adjust or stop the trial. Increased attendance would be a useful observation, but it would not by itself prove the new time caused the increase. Reminders, activity content and who was available may also have changed.
In the video, Marco describes a Monday meeting to review the previous week and plan ahead. The practical lesson is to put review into the team's routine. Choose a cadence that fits the decision; longer-term outcomes still need an appropriate follow-up period.
A first-pilot worksheet
| Item | Illustrative attendance pilot |
|---|---|
| Decision | Whether to offer an earlier session |
| Owner | Program manager |
| Evidence | Dated attendance, relevant participant feedback and staff notes |
| Uncertainty | Missing feedback; other reasons for absence |
| Proposed test | One site's earlier session after consultation |
| Review | Attendance coverage, participant experience and delivery burden |
| Decision record | Keep, adjust or stop, with reasons and a next check |
Use this prompt only with authorized records:
Review attendance and related feedback for [program, site and dates]. Use [attendance definition] and the eligible participant list. Show expected and attended counts, missing records and response coverage. List reported barriers with links to the exact supporting records. Separate participant statements, staff interpretations and your hypotheses. Include contradictory evidence. Do not invent reasons for absent participants or infer sensitive traits. Return questions for the program team to verify and possible small tests for human review. Do not make individual eligibility or safeguarding decisions.
6. Use the same evidence for different audiences
Staff need an action list. Participants need to know what happened in response to their feedback. A board needs choices, risks and resource implications. Funders need results and a clear account of what remains uncertain.
AI can help draft these versions from the same reviewed findings. A person checks each version, and each audience receives only the information it should see. A board summary usually does not need identifiable case notes. Keep the date, population and definitions consistent across the versions.
7. Expand when the first workflow works
Once attendance and follow-up are useful, consider the next related job: referrals, another program, membership or maintenance. Reuse relevant history and agreed definitions instead of starting a disconnected record each time.
Marco's account describes an expansion from impact measurement into operational work, including participation, staff work, supplies, maintenance and water management. It illustrates how needs can emerge during use. It does not establish a universal savings percentage or prove improved participant outcomes.
What should you measure in the pilot?
Keep three questions separate. Did AI perform the assigned task accurately enough after review? Did the complete process help staff act with less repeated work? Did the program change improve participants' experience or outcomes?
Record preparation, checking and correction work alongside any time saved. Track who is missing from the evidence. Ask what became easier and what new burden appeared. A fast summary is one output; a verified improvement for participants is a different result.
Where does Sopact fit?
A team can start with an attendance sheet, authorized notes and a shared action register. Repeated collection becomes harder when staff must match people, rebuild context and recode comments every cycle.
Sopact's connected collection and analysis approach can keep relevant records available together for recurring questions. Test it against the same requirements: source support, corrections, access controls, manageable routine changes and follow-up. People retain responsibility for interpretation and action. The benefit should be demonstrated in your pilot, including the work needed to maintain it.
Frequently asked questions
Where should a small nonprofit start with AI?
Choose one repeated question with a named owner and a reviewable answer. Attendance barriers, recurring maintenance issues or duplicated staff work can provide a bounded starting point. Use authorized information and define the review method before testing. Start small enough that the team can check errors and see what happens after acting.
Do we need to replace our CRM?
Not necessarily. Identify which jobs the existing system supports well and where context is lost between systems. Test the complete workflow before deciding whether to configure, connect or replace anything. Include the cost of maintaining those connections and protect the continuity of participant and operational records.
Can AI tell us why participants stop attending?
It can organize reported reasons and suggest questions to investigate. It cannot know the reasons of people who supplied no explanation, and an association is not proof of cause. Check the source, timing and coverage, then involve the relevant staff and participants in deciding what the pattern means.
What should staff verify in an AI answer?
Check the population, period, attendance definition, missing records and source support for each claim. Read contradictory accounts as well as the examples selected by the assistant. Recalculate important numbers. Confirm that the answer uses only permitted data and that an interpretation has not been presented as a participant's statement.
How do we know whether an operational change helped?
Decide what evidence you will review before making the change. Track delivery, participant experience and the relevant outcome separately. Compare compatible periods and explain other changes that might affect the result. Use the evidence to keep, revise or stop the change without overstating what the comparison proves.
What should we share with funders?
Explain the problem, what the team tried, what changed, what remains uncertain and what further support would enable. Show operational improvement alongside program results, with clear definitions and coverage. Use reviewed evidence and suitable access controls; a persuasive narrative should not expose private participant details or turn a trial into a proven impact claim.
Continue with Manage the Work, Connect Quantitative and Qualitative Data, Prepare Data Governance, and AI for Social Impact.