DonorConnect

Making a difference together

AI Policy & Safeguards

How DonorConnect uses AI responsibly to assist, not replace, nonprofit staff.

🤖 AI-Powered Feature: Donor Summaries

DonorConnect includes an AI-powered donor summary feature that helps nonprofit staff quickly understand donor relationships and personalize outreach. This feature is designed to augment human decision-making, not replace it.

How It Works:

Staff clicks the "AI Summary" button on a donor's profile page
Server validates the request and fetches donor data (organization-scoped)
A minimal summary payload is sent to OpenAI (no raw PII)
AI generates an 80-100 word actionable summary with risk assessment
Summary is displayed to staff with suggested next actions

📡 AI Model & API Used

Model	OpenAI GPT-4o-mini
API Endpoint	`POST /api/ai/summarize-donor`
Provider	OpenAI API (api.openai.com)
Max Tokens	200 tokens
Temperature	0.4 (low creativity, high consistency)

🛡️ How We Use AI Responsibly

✅ What We DO

Use AI for summaries and suggestions only
Keep API keys server-side only
Validate organization access before AI calls
Send minimal, aggregated data (not raw records)
Give users control via opt-in button
Return user-friendly error messages

❌ What We DON'T Do

Expose API keys to the client
Send full addresses, phone numbers, or SSNs
Allow AI to make decisions (only recommendations)
Train models on user data

📊 Data Sent to AI

We minimize data exposure by sending only aggregated metrics, never raw PII:

{
  "name": "John Doe",           // First + Last name only
  "status": "ACTIVE",           // Donor status enum
  "retentionRisk": "MEDIUM",    // Calculated risk level
  "totalAmount": 1250.00,       // Lifetime giving total
  "totalGifts": 5,              // Number of donations
  "lastGiftDate": "2025-12-15", // Most recent gift
  "notes": "Prefers email...",  // Staff notes for personalization
  "recentDonations": [          // Last 10 gifts only
    { "amount": 100, "date": "2025-12-15" }
  ]
}

NOT sent: Full address, phone number, email address, payment details, or raw donation records.

✍️ Prompt Engineering

We crafted prompts to generate concise, factual, and actionable summaries:

System Prompt:

"Be concise, factual, and actionable. Do not invent data."

User Prompt:

"You are a donor success assistant. Summarize this donor in 80-100 words max with risk assessment and suggested next action. Use any notes about the donor to personalize the recommendation."

Prompt Design Principles:

Length constraint (80-100 words) prevents verbose, unfocused output
"Do not invent data" instruction reduces hallucination risk
Actionability focus ensures output is useful for staff decisions
Low temperature (0.4) prioritizes consistency over creativity
Personalization via notes lets staff guide AI recommendations

🚀 How AI Improves DonorConnect

Without AI

Staff manually reviews donation history
Risk assessment is subjective and inconsistent
Personalized outreach takes significant time
New staff lack context on donor relationships

With AI

Instant donor summary with one click
Consistent risk assessment based on data
AI suggests specific next actions
New staff quickly understand donor context

Bottom line: AI helps small nonprofit teams act faster and more consistently, improving first-to-second gift conversion rates without requiring additional staff.

🔒 Security & Access Controls

✓
Server-side only: All AI calls happen on the server. API keys are never sent to the browser.
✓
Organization scoping: Donors are validated against the user's organization before any AI request. Users cannot summarize donors from other organizations.
✓
Session authentication: All AI endpoints require valid session authentication via HTTP-only cookies.
✓
Error handling: Failures return user-friendly messages; stack traces and API details are never exposed.
✓
No data retention: OpenAI API requests are transient; we do not store or log AI conversations.

Questions about our AI usage? Review our source code on GitHub.