Data rule for this cycle: Prompts receive only anonymized, aggregated data points. No individual scholar names, financial records, family details, or personally identifiable information enters any AI prompt. All data is verified against source records before the prompt is run.
Scope for this cycle: One donor update format (individual major donor) and one board report section (program outcomes summary). All other report types carry to future cycles after this pilot is measured and stable.
Program Policy Statement
AI-Assisted Donor & Board Reporting — We Will / We Will Not
✓ We Will
- Use AI to convert verified, anonymized program data into narrative first drafts for donor updates and board sections.
- Supply all data points to the AI ourselves from verified source records before any prompt is run.
- Require staff and program director review of every AI draft before it leaves the organization.
- Personalize donor updates with specific relationship context added by development staff after AI drafts the narrative baseline.
- Log which reports used AI assistance and retain the data input sheet as part of the documentation record.
- Measure time savings and error rate against last cycle's reports before expanding to additional formats.
✕ We Will Not
- Include individual scholar names, family financial information, or any PII in any AI prompt.
- Allow AI to generate or extrapolate data. Every number in an AI draft must trace to a verified source record.
- Send any AI-drafted report without staff and director review and explicit approval.
- Use AI to draft reports for high-stakes, legally sensitive, or complex narrative situations (governance disputes, audit findings, funding shortfalls).
- Expand to additional report types (annual reports, grant reports, federal filings) until this pilot is measured.
1
Weeks 1–2
Planning: Audit Reports, Map Data Sources, and Define Scope
Sprint goal: Leave Weeks 1–2 with a report type inventory, a verified data source map, a prompt-ready data template for each in-scope report, and a policy statement on file.
Task 1 — Report Type Inventory and Time Audit
Map every donor and board report your program produces. Identify where time is lost and where AI can convert structured data into narrative with the lowest risk.
Donor-Facing Reports
- Individual major donor impact update (in scope for this cycle)
- Annual fund appeal letter
- Named scholarship report to family
- Grant progress report (funder-specific format)
- Acknowledgment and stewardship letter
Board-Facing Reports
- Program outcomes summary section (in scope for this cycle)
- Full board meeting packet
- Annual report narrative
- Strategic dashboard commentary
- Audit summary narrative
| Report Type | Frequency | Avg. Draft Time (hrs) | Primary Challenge | AI Pilot Candidate? |
|---|---|---|---|---|
| Individual donor impact update | Per cycle | Fill in | Personalizing from generic program data | YES — Start here |
| Board outcomes summary section | Quarterly | Fill in | Converting data tables into readable narrative | YES — Start here |
| Named scholarship report | Per cycle | Fill in | Scholar-specific narrative without PII risk | Cycle 2 |
| Grant progress report | Per grant | Fill in | Funder-specific format; compliance language | Cycle 3 |
| Audit summary narrative | Annual | Fill in | Legal sensitivity; board governance risk | Human-only |
Task 2 — Data Source Map and Verification Protocol
Before any prompt work begins, map exactly which data points go into each in-scope report and where they come from. Every data point must be verifiable against a source record before it enters a prompt.
- List every data point your standard donor impact update contains (total awards, average award, cohort demographics in aggregate, outcomes data such as enrollment rates or GPA averages, quotes or vignettes if anonymized).
- For each data point, identify the source system: scholarship management platform, registrar data, survey results, or program records.
- Assign a verification owner (the staff member who confirms accuracy before each report cycle).
- Create a prompt-ready data input template: a short, structured form that staff fill in with verified numbers before running any prompt. This form is the only input the AI receives.
- Have the program director review the data input template before the first prompt run. Any field that cannot be verified from a source system is removed from the template.
Critical step: The AI draft is only as accurate as the data you provide. If the data input template has errors, the AI will amplify them. Verify first, then prompt.
Task 3 — Prompt-Ready Data Input Template (Donor Update)
Staff fill this template with verified data before running any prompt. This is the only source of program-specific information in the prompt. Keep it short and factual.
| Data Field | Example Input | Source System | Verified By |
|---|---|---|---|
| Award cycle year | 2025–26 | Program records | |
| Number of scholars supported | 42 | Scholarship platform | |
| Total amount awarded | $185,000 | Financial records | |
| Average award per scholar | $4,405 | Calculated from above | |
| Cohort demographic summary (aggregate %) | 68% first-generation, 54% rural | Application survey | |
| Outcome data point 1 | 92% enrolled full-time (prior cohort) | Registrar / follow-up survey | |
| Outcome data point 2 | Fill in or write "none verified" | ||
| Donor-specific giving history | [DO NOT INCLUDE IN AI PROMPT — add manually post-draft] | Development CRM | |
| Approved program mission statement | [PASTE APPROVED TEXT] | Organization website | |
| Tone / relationship note | Long-term major donor; personal; avoid jargon | Development staff |
The "Donor-specific giving history" row is marked DO NOT INCLUDE in the AI prompt. This is added by development staff manually after the AI draft is reviewed. It should never be pasted into any AI tool.
Sprint 1 Deliverable Checklist
Report type inventory with time log
Data source map (verified)
Prompt-ready data input template
Data handling agreement
Policy statement (signed)
Director sign-off on data template
2
Weeks 3–4
Building: Prompts, Test Drafts, and Calibration
Sprint goal: Produce a tested, three-tier prompt set for both in-scope report types, a calibration log, a post-draft verification checklist, and a one-page workflow reference for development staff.
Task 4 — Calibration Protocol (Test Drafts)
Before using the prompts on real reports, run 3–5 test drafts using last cycle's verified data (already public or safely aggregated). Compare outputs to the actual reports sent last cycle.
- Use last cycle's verified data to fill in the prompt-ready data input template. All figures must already be public or anonymized.
- Run the Fast Draft prompt. Compare the output to the actual donor update sent last cycle.
- Document differences: where did the AI add language not in the data input? Where did it miss key program language? Where did the tone drift?
- Each documented failure becomes a constraint added to the High-Accuracy prompt (e.g., "Do not use the word 'transform' unless it appears in the approved mission statement.").
- Run the revised prompt. Calibration is complete when development staff can get from prompt output to a send-ready draft in under 15 minutes of editing.
Calibration Log Template
| Test Draft # | Report Type | Invented Data? | Tone Drift? | Mission Language Accurate? | Edit Time (min) | Action Taken |
|---|---|---|---|---|---|---|
| 1 | Donor update | |||||
| 2 | Donor update | |||||
| 3 | Board outcomes section | |||||
| 4 | Board outcomes section |
Sprint 2 Deliverable Checklist
Three-tier prompt set (both report types)
Calibration log (completed)
Post-draft verification checklist
One-page workflow reference
Director sign-off on prompt version
3
Weeks 5–6
Review & Wrap-Up: Measure, Audit for Accuracy, and Scope Next Cycle
Sprint goal: Produce a before/after time log, a factual accuracy audit across the live batch, and a next-cycle brief. Identify which report types are strong candidates for Cycle 5.
Factual Accuracy Audit Protocol
Pull 5–10 AI-drafted reports that were reviewed and sent. Verify each data claim against the original source record.
- CheckEvery number in the sent report traces to the verified data input template. No invented figures.
- CheckNo individual scholar names, specific school names, or personal details appeared in any sent report.
- CheckMission language matches the approved organization statement. No paraphrased variants were introduced by AI and left uncorrected.
- CheckOutcome claims (enrollment rates, GPA data, etc.) are from the correct cohort year and labeled correctly.
- NoteWere any donor-relationship personalizations missed (i.e., staff forgot to add the manual CRM layer)?
Tone and Relationship Audit
Report inaccuracies are the highest risk, but tone misalignment damages donor relationships. Audit a sample against each recipient relationship type.
- CheckWas the tone appropriate to the relationship level (major donor vs. annual fund vs. institutional funder)?
- CheckDid any language imply outcomes or impact the program cannot verify or substantiate?
- CheckWere any phrases overly generic or interchangeable with reports from other organizations (signal that AI filled space without your program's data)?
- NoteDid board members or donors respond with clarifying questions about specific claims? Log these as prompt-improvement signals.
Carry-forward rule: Any report where the AI invented a data point that staff missed in review must be documented and escalated before expanding to additional report formats.
Sprint 3 Deliverable Checklist
Before/after time log
Factual accuracy audit summary
Error log (invented data, tone drift)
Revised prompt pack (if needed)
Next milestone brief
Full Prompt Pack — Donor Impact Updates & Board Outcomes Section
All prompts receive only the verified data input template as input. No names, no CRM data, no unverified figures. The data template is filled in and reviewed by staff before any prompt is run.
Good — Fast Draft: Donor impact update from verified data block
You are a development writer for a nonprofit scholarship program. Your task: Draft a donor impact update using only the verified data below. Convert the data into 2–3 warm, readable paragraphs that communicate program impact to a major donor. Verified program data for this update: [PASTE YOUR COMPLETED DATA INPUT TEMPLATE HERE] Writing requirements: - Tone: warm, personal, grateful — not transactional or corporate - Length: 250–350 words - Format: 3 short paragraphs (program context, impact data narrative, forward-looking close) - Use only data provided above; do not extrapolate, estimate, or invent outcomes - Do not include: donor name, giving amount, or relationship history (these will be added by staff) Output: One complete narrative draft, ready for staff to personalize with relationship context.
Use for: standard major donor impact update. Staff add the personalization layer (relationship context, specific gift acknowledgment) manually before review and send.
Better — High-Accuracy: Adds self-check, flags invented data, verifies mission language
You are a development writer for a nonprofit scholarship program. Your task: Draft a donor impact update narrative from the verified data block below. Verified program data: [PASTE COMPLETED DATA INPUT TEMPLATE] Tone context: [e.g., "Long-term major donor, conversational, mission-connected" OR "Institutional funder, formal, outcomes-focused"] Writing requirements: - Tone: warm, personal, grateful - Length: 250–350 words - Format: 3 paragraphs (context, impact, forward close) - Use only data I have provided; cite only figures that appear in the data block above - If you are uncertain whether a claim is supported by the data, write [STAFF: VERIFY THIS] inline Self-check before outputting: 1. Does any sentence include a figure not in the data block? Remove it. 2. Does any sentence include a scholar name or any PII? Remove it. 3. Does the mission language match the approved statement in the data block? 4. Is the tone consistent with the tone context provided? Output: Draft narrative followed by a list of any [STAFF: VERIFY] flags with the specific concern for each.
Use for: standard cycle updates where accuracy verification matters. Review all [STAFF: VERIFY] flags before the personalization step.
Best — Governed Workflow: Full audit trail for board and high-stakes reports
You are a development writer for a nonprofit scholarship program. This is a governed workflow. Follow all steps in order. STEP 1 — Acknowledge scope: Confirm: (a) no PII or individual scholar data is present in this prompt, (b) your output is a narrative draft for staff review only — not a final document, (c) you will flag any claim you cannot trace to the data block rather than inventing language. STEP 2 — Draft the narrative using only this verified data: [PASTE COMPLETED DATA INPUT TEMPLATE] Report type: [DONOR IMPACT UPDATE or BOARD OUTCOMES SECTION] Audience: [MAJOR DONOR / BOARD MEMBER / OTHER — describe relationship briefly] Length and format: [e.g., "3 paragraphs, 300 words, warm and personal" or "4 bullet points + 1 summary paragraph, formal, outcomes-focused"] STEP 3 — Self-audit before outputting: - [ ] Every number in the draft traces to a field in the data block above? - [ ] No scholar names, family details, or individual PII present? - [ ] Mission language matches the approved statement provided? - [ ] No outcome claimed that is not in the verified data? - [ ] [STAFF: VERIFY THIS] inserted wherever any claim is uncertain? STEP 4 — Output format: A. Self-audit results (one line per check above) B. Complete narrative draft C. List of [STAFF: VERIFY] flags with the specific data gap for each D. Personalization reminder: list the 2–3 fields staff should add manually from the CRM before finalizing
Use for: board reports, high-stakes donor updates, and any time full audit documentation is required. Retain the self-audit output as part of the report documentation record.
Sample AI Draft Output — Donor Impact Update
This is what a completed, staff-reviewed AI draft should look like before personalization. All figures trace to the verified data input template. The [PERSONALIZATION] markers are removed and replaced by development staff from CRM data before the letter is sent.
Sample Donor Impact Update — AI Draft (Post-Review, Pre-Personalization)
Staff-reviewed draft
[PERSONALIZATION: Opening with specific acknowledgment of donor's relationship and history — added by development staff from CRM]
The 2025–26 award cycle marked another year of meaningful reach for [PROGRAM NAME]. Forty-two scholars received awards totaling $185,000, with an average award of $4,405 per student. Nearly seven in ten scholars are the first in their families to pursue higher education, and more than half come from rural communities where access to scholarship support remains limited.
All figures above are from the verified 2025–26 program records. Verify before finalizing: cohort demographics are from the application survey (not registrar-confirmed).
The outcomes from prior cohorts continue to affirm the program's core purpose. Ninety-two percent of scholars from the previous award cycle enrolled full-time in their first year, a figure that speaks to both the students' commitment and the stability that scholarship support provides during a critical transition.
[PERSONALIZATION: Closing that connects donor's specific contribution to program outcomes and expresses specific gratitude for their partnership — added by development staff from CRM]
Staff flags: One [STAFF: VERIFY] note inserted at cohort demographics sentence. Data input template lists this as "Application survey" source — confirm with program coordinator that this figure is current before finalizing. No other invented data found.
Staff Verification Checklist — Before Any Report Is Sent
Required for every AI-assisted report during the pilot. Staff initials and date required before the report moves to the director approval queue.
Section 1 — Data Accuracy
- RequiredEvery figure in the report has been traced back to the verified data input template and confirmed accurate.
- RequiredNo individual scholar names, family financial data, or PII appears anywhere in the report.
- RequiredAll [STAFF: VERIFY] flags from the governed workflow have been resolved or corrected.
- ReviewMission language matches the current approved organizational statement.
- ReviewOutcome data is labeled with the correct cohort year (e.g., "prior cohort," "2024–25 cycle").
Section 2 — Personalization and Relationship
- RequiredDonor-specific relationship context has been added manually (not generated by AI) from the CRM.
- RequiredTone is appropriate to this specific donor relationship (not a generic template voice).
- ReviewNo language implies outcomes or projections the program cannot substantiate.
- ReviewCall to action or next steps (if included) are accurate for this donor's stage in the relationship.
- LogReport logged as AI-assisted in the development tracking sheet for end-of-cycle audit.
Section 3 — Director Approval Sign-Off
- RequiredDevelopment staff reviewer: _______________ Date: _______________
- RequiredProgram director approval: _______________ Date: _______________
- BoardFor board reports: board liaison reviewed and confirmed accuracy of board-specific framing before distribution.
Measurement Framework
Primary Metric
Draft Time
Hours from data verification to send-ready report. Track before and after across the full batch.
Accuracy Metric
Error Rate
Count of invented or unverified data points found during staff review. Target: zero in any sent report.
Relationship Metric
Follow-up Qs
Clarifying questions from donors or board members after receiving the report. Unexpected questions signal AI fill-in or accuracy issues.