We are moving past simple “if this, then that” logic. The modern stack isn’t about connecting apps; it’s about deploying autonomous agents that think, route, and execute complex labor without you.
If you are still manually categorizing emails, scraping leads by hand, or hiring VAs for data entry, you are bleeding efficiency.
This protocol breaks down the n8n platform—the open-source, node-based automation tool that eats Zapier for breakfast. We are building 8 production-ready workflows, ranging from simple lead responders to complex RAG (Retrieval-Augmented Generation) systems.
Strap in. We are building the infrastructure for a self-driving business.
Phase 1: The Foundations (Lead Response Protocol)
Objective: Reduce lead response time to <60 seconds to quadruple conversion rates.
The Logic:
Trigger: A form submission (using n8n's native form trigger).
Database: Log the lead in Google Sheets.
Alert: Notify the internal team.
Logic: Filter based on budget (High Value vs. Low Value).
Action: Send dynamic email responses.Step-by-Step Configuration:
The Trigger Node:
Add the "On Form Submission" node.
Config: Create fields for Full Name, Email, Service Type (Dropdown: SEO, Ads), and Budget (Number).
Test: Click "Execute Node" and submit a dummy entry to capture JSON data.
Google Sheets Logging:
Add Google Sheets node -> Action: Append Row.
Mapping: Drag the JSON output from the Form node (Name, Email, Budget) into the Sheet columns.
Function: Use the expression {{ $now }} in the Date column to timestamp entries automatically.
The Filter (Profitability Gate):
Add a Filter node.
Condition: Set Budget >= 1000.
Result: Only high-ticket leads pass this node. Low-ticket leads are discarded or routed to a "Downsell" path.
The Split Logic (If/Switch):
Add a Switch node.
Routing: Route based on Service Type.
Output 0: SEO
Output 1: Ads
This splits the workflow into two parallel paths for hyper-personalized email drafting.
Gmail Dispatch:
Add Gmail node -> Action: Send Email.
Subject: "Re: Your [Service Type] Inquiry."
Body: Use Expression Mode. "Hi {{ $json["name"] }}, I saw you're interested in {{ $json["service"] }}..."
Optimization: Use {{ $json["name"].split(" ")[0] }} in the expression editor to strip the last name and only use the first name for a natural tone.Phase 2: Inbox Zero (AI Email Classifier)
Objective: Use an LLM to read, categorize, and archive emails automatically.
The Logic:
Trigger: Gmail (Email Received).
Intelligence: OpenAI (Analyze sentiment/intent).
Routing: Switch Node (Personal, Spam, Sales, Urgent).
Action: Apply Labels/Archive.Step-by-Step Configuration:
Gmail Trigger:
Event: On Message Received.
Polling: Set to 10-30 minutes to save execution credits (polling every minute burns resources).
Simplify: Toggle OFF to get the full raw text body.
The AI Brain (OpenAI Node):
Model: gpt-4o-mini (Fast and cheap).
System Prompt: "You are an email sorter. Categorize the incoming email into one of these buckets: [Promotions, Social, Personal, Sales]. Output ONLY the category name."
User Message: Map the {{ $json["snippet"] }} or {{ $json["text"] }} from the Gmail trigger.
The Router (Switch Node):
Rules: Set routing rules based on the OpenAI output string.
Rule 1: If output contains "Promotions" -> Output 0.
Rule 2: If output contains "Personal" -> Output 1.
Automated Actions:
Promotions Path: Connect a Gmail node -> Action: Add Label ("Low Priority") -> Action: Mark as Read.
Personal Path: Connect a Gmail node -> Action: Create Draft.
Draft Prompt: Use a second OpenAI node to write a reply. "Draft a polite response to {{ $json["from"] }}. Sign off as Jonno."Phase 3: The Lead Scraper (Apify + Google Maps)
Objective: Scrape 1,000+ targeted business leads from Google Maps daily.
The Logic:
Trigger: Schedule (Every Morning).
Scraper: Apify (Google Maps Scraper Actor).
Storage: Google Sheets.
Optimization: Loop and Aggregation to handle rate limits.Step-by-Step Configuration:
The Trigger:
Add Schedule node. Set to 8:00 AM daily.
Apify Integration:
Add Apify node.
Action: Run Actor.
Actor ID: Search for the Google Maps Scraper (Compass).
Input JSON: Paste the search configuration (e.g., {"searchStrings": ["Plumbers in Toronto"], "maxCrawledPlaces": 10}).
Data Cleaning:
The output will be a massive JSON array. You must map specific fields (Name, Phone, Website, Review Score).
The Loop (Rate Limit Protection):
Problem: Sending 500 rows to Google Sheets instantly will crash the API.
Solution: Add a Split In Batches (Loop) node. Set batch size to 10.
Add a Wait node (2 seconds) inside the loop.
Google Sheets Update:
Inside the loop, add Google Sheets -> Action: Add or Update Row.
Key: Match based on "Website" or "Phone Number" to avoid duplicates.Phase 4: The AI Accountant (Telegram Receipt Parser)
Objective: Snap a photo of a receipt, have AI extract line items, and update your P&L sheet.
The Logic:
Trigger: Telegram (On Message).
Router: Is it an Image or PDF?
Extraction: OpenAI Vision (Image) or Text Parser (PDF).
Structuring: Force JSON output.
Storage: Google Sheets (Line Item Splits).Step-by-Step Configuration:
Telegram Setup:
Use BotFather to create a bot. Paste the API Token into n8n credentials.
Trigger: On Message (Listen for files/photos).
File Handling:
The trigger gives a file_id. You must add a Telegram node -> Action: Get File to download the binary data.
AI Vision Extraction:
Add OpenAI node -> Model: gpt-4o.
Input: Binary File.
System Prompt: "Extract all line items from this receipt. Return ONLY JSON format: [{item: 'name', price: 10.00}]."
Critical: Enable "JSON Mode" in OpenAI settings to ensure the code doesn't break.
The Split Out Node:
Since one receipt has multiple items, use the Item Lists node -> Split Out Items.
Field to split: line_items.
This turns one receipt object into 10 separate rows for Google Sheets.
Database Commit:
Map the split items into Google Sheets columns: Item Name, Cost, Date, Merchant.Phase 5: Hyper-Agents (Multi-Agent Systems)
Objective: A “Manager” Agent that delegates tasks to “Worker” Agents (Calendar management, Email management).
The Logic:
Interface: Telegram Chat.
Brain: Manager AI Agent.
Tools: Sub-workflows (Worker Agents).Step-by-Step Configuration:
The Manager Agent:
Add AI Agent node.
Tools: Instead of giving it direct API access, we create Workflow Tools.
Tool 1: "Calendar Worker" (Triggered by another workflow).
Tool 2: "Gmail Worker".
Creating the Sub-Workflow (Calendar Worker):
Create a new n8n workflow.
Trigger: Execute Workflow Trigger.
Inputs: Define JSON inputs: action (create/delete), date, title.
Logic: Use a Switch node. If action = create -> Google Calendar Node (Create Event).
Linking:
Back in the Manager Agent, select "Call Workflow" tool. Select the "Calendar Worker" workflow.
System Prompt: "You are a personal assistant. If the user asks to book a meeting, call the Calendar Worker tool with the correct date and title."
Memory Management:
Add Window Buffer Memory. Connect it to the Agent.
Session ID: Use the Telegram chat_id. This ensures the bot remembers context ("Change that meeting to 4 PM").Phase 6: The RAG System (Company Knowledge Base)
Objective: A chatbot that answers questions based only on your internal PDF documents.
The Logic:
Ingestion: Google Drive -> PDF Reader -> Vector Store (Pinecone).
Retrieval: Chat Interface -> Vector Search -> AI Response.Step-by-Step Configuration:
Vector Database Setup (Pinecone):
Create a free Pinecone index (Dimensions: 1536, Metric: Cosine).
In n8n, use Pinecone node -> Operation: Upsert.
Data Pipeline:
Google Drive (Download File) -> Read PDF (Extract Text).
Recursive Character Text Splitter: Chunk size 500, overlap 50. (Crucial: Don't feed the whole PDF at once; chunk it so the AI can find specific paragraphs).
Embeddings: Connect OpenAI Embeddings node (text-embedding-3-small) to the Pinecone node.
The Chat Agent:
Trigger: Chat Trigger.
AI Agent node connected to Vector Store Tool.
Tool Config: Operation Retrieve. Limit results to 4.
System Prompt: "You are a support bot. Answer queries using ONLY the context provided by the Vector Store tool. If the answer isn't there, say you don't know."Phase 7: The SaaS Backend (Lovable + n8n)
Objective: Build a functional Micro-SaaS that generates social media posts from a URL.
The Logic:
Frontend: Lovable.dev (Generates the UI).
Backend: n8n Webhook.
Processing: Scrape URL -> AI Copywriting -> AI Image Gen.
Response: Return JSON to frontend.Step-by-Step Configuration:
The Webhook:
Add Webhook node. Method: POST.
Copy the Production URL.
The Frontend (Lovable):
Prompt Lovable: "Build a form with fields for Company Name and URL. On submit, POST data to [Your_n8n_Webhook_URL]."
The Processing Chain:
HTTP Request: GET the user's URL.
HTML to Text: Strip the HTML tags to save tokens.
OpenAI (Copy): "Write a LinkedIn post based on this website content."
DALL-E 3: "Generate a corporate Memphis style illustration for this post."
The Response:
Respond to Webhook node.
Format: JSON.
Body: { "post": "Generated text...", "image_url": "https://..." }
Critical: Ensure CORS is handled if running from a browser, or use a proxy.Technical Glossary & Troubleshooting
JSON Errors: If you see "Invalid JSON," you likely have line breaks in your string. Use JSON.stringify() in the expression editor to sanitize text before sending it.
Split Out vs. Aggregate: Use Split Out to turn a list (Array) into individual items (to process them one by one). Use Aggregate to bundle them back up (e.g., to send one summary email instead of 50 individual ones).
Cron Expressions: For schedule triggers, use crontab.guru to calculate the exact timing (e.g., 0 9 * * 1 for every Monday at 9 AM).
Execution Limits: On the free/starter plans, be careful with infinite loops. Always put a "Wait" node in loops to avoid hitting API rate limits.
