CRM Automation & Donor Management

A recurring pain point: shluchim spend hours on data entry, lapsed-donor follow-ups, and contact enrichment. The chat has tested several patterns.

The pain

"Update my CRM. I can forward it email, voice notes etc. and it would either make a new contact or add a note to a contact. Also if I could send it a voice note to schedule a follow up in my calendar."

"Analyze your CRM. See which donors lapsed and feed you 3 each day. Check when you last called them in your phone log or texts or WhatsApp or email."

Both bubble to the top whenever the group is asked "if you had an AI agent that could do anything for you, what would it do?"

The CRMs in play

CRM	Notes from the chat
ChabadOne (Salesforce-backed)	The Chabad-default. AI integration via Salesforce MCP, reported as easy to set up. Being tested actively.
Hecher CRM	Custom-built with Claude MCP integration. An 80-shliach beta. Voice notes, transcribe (Whisper), then Claude turns them into emails, calendar items, and CRM updates.
Little Green Light (LGL)	Mainstream nonprofit CRM. An AI agent that self-learns features via Nanoclaw is in heavy use.
CiviCRM / Chabad Suite	API exists; a member connected it to Base44 to push/pull data.
MyShul	Has its own AI report-bot: ask "give me a list of all members in a given zip code" and it produces it.
Attio	Modern CRM. An agent is used for prospect research and contact enrichment.

Patterns that work

1. Agent-on-CRM (Claude Dispatch / Cowork / Nanoclaw)

Claude Dispatch acts like a VA clicking through LGL, adds donations, runs reports, drafts follow-ups. It doesn't use the LGL API; it uses browser automation:

"It's like an agent working on your computer, in your browser. Pretty cool. Like having a VA, but without a brain, so doesn't make low-IQ mistakes."

Nanoclaw is the safer version: ex-JLI / ex-IDF 8200 founders, Docker-sandboxed.

2. MCP / API into vetted backends

When the CRM has an API or MCP server, you skip browser automation:

• Salesforce MCP → Claude/ChatGPT → ChabadOne data.
• Hecher MCP → a member's bespoke setup.
• CiviCRM API → Base44 (no-code) frontend.

This is the engineering path. Slower to set up; reliable in production.

3. Voice note to action-item pipeline

A common request, partially solved:

1. Record a voice note (WhatsApp or built-in voice memo).
2. Transcribe with Sofer.ai or Whisper (free open-source model).
3. Feed the transcript plus a system prompt to Claude or ChatGPT.
4. System prompt: "Extract: new contacts to add, calendar follow-ups, email drafts. Output JSON."
5. Push to the CRM via API.

The Hecher CRM, a member's build, has this end-to-end.

4. Donor research / enrichment

Use cases the group has shipped:

• Public-philanthropist lookup. Claude Cowork built spreadsheets of publicly listed top Jewish donors per city. One member said it "saved me a few hours."
• Email-finding via Sheets formula. Concatenate everything you know about a person into one cell, use the =AI(...) Gemini Workspace formula, drag down. Retry failures. Caution: avoid feeding donor PII into consumer AI; prefer non-identifying fields or aggregates.
• WealthEngine for prospect scoring.
• Hatch.ai, Lubavitch-owned prospect research; some users find it "more hatch and less ai."

What doesn't work

• Vibe-coded CRMs with login + payment. Repeatedly warned against; see Vibe Coding. Keep banking out of AI-built systems.
• Web-UI click-bots on flaky pages. Claude Cowork is good but fragile; CRM UI changes break runs.
• Ethnicity-based marketing lists. A real legal restriction (Data Axle and similar), not just an AI guardrail.

Open questions

• A CRM template that 200 shluchim could install without each one becoming a security incident. The community keeps proposing; no one has shipped it.
• Email plus voice note to CRM as a packaged install instead of a bespoke build per shliach.
• WhatsApp data mining. A Selenium-based exporter is the only working solution; encrypted backups block direct access.

Related

• Claude, the Dispatch / Cowork / MCP path.
• Vibe Coding
• Fundraising