Voice-Driven CRM — Closing the Loop Between Thinking and Data

The Insight

Don Back shared a workflow he built overnight:

“Last night I found myself with some free time so I thought I’d finish the next step of my CRM project. I wrote a custom GPT to act as an interpreter/agent to take an oral database update, pass it to Google applications script and update the Google Sheet that I’m using as a database. The last step I’m working on is finalizing an apple script so that A ‘Hey Siri’ command will trigger the custom GPT so that it all works verbally.”

The result: a fully voice-driven CRM update loop. You speak a note about a client interaction, Siri triggers the custom GPT, the GPT interprets the update, Google Apps Script formats and writes it to a Google Sheet acting as a live database.

Why It Matters

CRM data quality degrades not because people don’t want to update it — it’s because the friction of updating it always loses to whatever else is happening in the moment. Voice removes the primary friction point (keyboard + navigation) and places the capture action at the exact moment the thought is fresh.

The architecture is significant: it demonstrates that complex multi-step integrations (voice → LLM → scripting → database) are now achievable by non-engineers in a single session. The custom GPT acts as the natural language interpreter that eliminates the need for structured voice commands.

The Architecture

"Hey Siri" (trigger)
    → AppleScript (local automation bridge)
    → Custom GPT (natural language → structured update)
    → Google Apps Script (execution layer)
    → Google Sheet (database)

Each layer has a single responsibility:

• AppleScript: trigger on voice keyword, route to GPT
• Custom GPT: interpret free-form spoken update into structured data
• Google Apps Script: write the structured data to the correct row/column
• Google Sheet: persistent, searchable, sharable database

Connection to PowerUp Themes

This is a concrete demonstration of Insight - Build Tiny Tools That Remove Real Friction — the friction being removed is the gap between “I just finished a client call and have a clear thought” and “that thought is captured somewhere it can be retrieved and acted on.”

It also connects to Insight - Codify Your Judgment Into Skills, Not Just Prompts: the custom GPT here is essentially a codified judgment about what constitutes a useful CRM entry and how to structure it.

Application for Coaches

• The same architecture works for any note-taking workflow: morning reflections, client session notes, follow-up commitments
• Google Sheets as the database layer is deliberately simple — no backend, no API keys, viewable by any collaborator
• The custom GPT prompt can encode CRM-specific rules: what fields to update, how to handle ambiguous references, what to flag for follow-up
• The voice trigger removes the “open app, find contact, click, type” sequence that makes CRM hygiene feel like overhead rather than a natural part of a client conversation

Replication Path

1. Build a Google Sheet with the fields you need (contact, date, note, next action)
2. Write a custom GPT prompt that parses spoken updates into those fields
3. Build a Google Apps Script that accepts structured input and writes to the sheet
4. Build an AppleScript that sends voice input to the GPT and routes the output to Apps Script
5. Assign the AppleScript to a Siri shortcut