Auto-resolve support tickets and escalate the rest
I built an n8n workflow that acts as a first-line AI support agent. Task: I built an n8n workflow that acts as a first-line AI support agent. It receives customer tickets via webhook, tries to resolve them using your knowledge base and order history, and either sends an auto-reply or escalates to the right human team — all automatically. How it works Customer submits a ticket via webhook (works with any website form, chat widget, or API call) Customer submits a ticket via webhook (works with any website form, chat widget, or API call) n8n parses the ticket and runs two lookups in parallel: knowledge base search + order history n8n parses the ticket and runs two lookups in parallel: knowledge base search + order history GPT-4o-mini analyzes everything — classifies the category (billing, technical, shipping, etc.), assigns priority, detects sentiment, and writes a response GPT-4o-mini analyzes everything — classifies the category (billing, technical, shipping, etc.), assigns priority, detects sentiment, and writes a response If AI can resolve it → sends a branded HTML email reply to the customer, logs to Google Sheets, posts to Slack If AI can resolve it → sends a branded HTML email reply to the customer, logs to Google Sheets, posts to Slack If it can’t → escalates to the right Slack channel (#billing-support, #tech-support, #security-urgent) with the AI’s draft response and internal notes so the human agent can respond faster If it can’t → escalates to the right Slack channel (#billing-support, #tech-support, #security-urgent) with the AI’s draft response and internal notes so the human agent can respond faster Smart escalation The AI doesn’t just dump everything into one queue. It routes based on the issue: Billing/refund issues → always escalated to #billing-support Billing/refund issues → always escalated to #billing-support Account security → immediately escalated to #security-urgent Account security → immediately escalated to #security-urgent Technical issues beyond the KB → #tech-support Technical issues beyond the KB → #tech-support Everything else → #support-escalations Everything else → #support-escalations Each escalation includes the AI’s draft response, confidence score, and internal notes. What you need n8n (self-hosted or cloud) n8n (self-hosted or cloud) OpenAI API key OpenAI API key SMTP credentials SMTP credentials Google Sheets + Slack bot token Google Sheets + Slack bot token Optionally: a knowledge base API and order/CRM API Optionally: a knowledge base API and order/CRM API Cost With GPT-4o-mini, it’s about $0.90/month for 50 tickets/day. Very cheap to run. Get it Full workflow JSON, setup instructions, and detailed docs: https://github.com/ivansiyanko/n8n-support-ai Just import the JSON into n8n, set up your credentials, and activate. Happy to answer any questions or hear suggestions for improvements! Related topics; tools: n8n. Human role: The original post does not say how a person stayed involved.; limitations: Requires n8n, OpenAI API, SMTP, Google Sheets, and Slack credentials. Optional integrations include knowledge base and CRM APIs. The source does not specify error handling for API failures or hallucination safeguards. Human involvement is implied in escalation but not detailed in the workflow logic..
Tools
Published on Agent Work Atlas:
Overview
I built an n8n workflow that acts as a first-line AI support agent.
Workflow
I built an n8n workflow that acts as a first-line AI support agent. It receives customer tickets via webhook, tries to resolve them using your knowledge base and order history, and either sends an auto-reply or escalates to the right human team — all automatically. How it works Customer submits a ticket via webhook (works with any website form, chat widget, or API call) Customer submits a ticket via webhook (works with any website form, chat widget, or API call) n8n parses the ticket and runs two lookups in parallel: knowledge base search + order history n8n parses the ticket and runs two lookups in parallel: knowledge base search + order history GPT-4o-mini analyzes everything — classifies the category (billing, technical, shipping, etc.), assigns priority, detects sentiment, and writes a response GPT-4o-mini analyzes everything — classifies the category (billing, technical, shipping, etc.), assigns priority, detects sentiment, and writes a response If AI can resolve it → sends a branded HTML email reply to the customer, logs to Google Sheets, posts to Slack If AI can resolve it → sends a branded HTML email reply to the customer, logs to Google Sheets, posts to Slack If it can’t → escalates to the right Slack channel (#billing-support, #tech-support, #security-urgent) with the AI’s draft response and internal notes so the human agent can respond faster If it can’t → escalates to the right Slack channel (#billing-support, #tech-support, #security-urgent) with the AI’s draft response and internal notes so the human agent can respond faster Smart escalation The AI doesn’t just dump everything into one queue. It routes based on the issue: Billing/refund issues → always escalated to #billing-support Billing/refund issues → always escalated to #billing-support Account security → immediately escalated to #security-urgent Account security → immediately escalated to #security-urgent Technical issues beyond the KB → #tech-support Technical issues beyond the KB → #tech-support Everything else → #support-escalations Everything else → #support-escalations Each escalation includes the AI’s draft response, confidence score, and internal notes. What you need n8n (self-hosted or cloud) n8n (self-hosted or cloud) OpenAI API key OpenAI API key SMTP credentials SMTP credentials Google Sheets + Slack bot token Google Sheets + Slack bot token Optionally: a knowledge base API and order/CRM API Optionally: a knowledge base API and order/CRM API Cost With GPT-4o-mini, it’s about $0.90/month for 50 tickets/day. Very cheap to run. Get it Full workflow JSON, setup instructions, and detailed docs: https://github.com/ivansiyanko/n8n-support-ai Just import the JSON into n8n, set up your credentials, and activate. Happy to answer any questions or hear suggestions for improvements! Related topics
Reported result
The source does not report measured outcomes, success rates, or resolution times. It only claims the workflow is operational and cost-effective at approximately $0.90/month for 50 tickets/day. No data on actual customer satisfaction or error rates is provided.
Human role
The original post does not say how a person stayed involved.
Limitations
Requires n8n, OpenAI API, SMTP, Google Sheets, and Slack credentials. Optional integrations include knowledge base and CRM APIs. The source does not specify error handling for API failures or hallucination safeguards. Human involvement is implied in escalation but not detailed in the workflow logic.
Try it
Act as an automation engineer. I need to replicate the n8n AI support agent workflow described in the source. Inputs: [Replace with your n8n instance URL], [OpenAI API Key], [SMTP Credentials], [Slack Bot Token], [Google Sheets ID], [Knowledge Base API Endpoint], [Order History API Endpoint]. Steps: 1. Create an n8n workflow starting with a Webhook node to receive ticket JSON. 2. Add parallel HTTP Request nodes to fetch data from the Knowledge Base and Order History APIs. 3. Insert an OpenAI node using GPT-4o-mini to classify the ticket (billing, technical, etc.), assign priority, detect sentiment, and draft a response. 4. Add an IF node to check if the AI confidence score is high enough to auto-resolve. 5. If true: Send an HTML email via SMTP, log to Google Sheets, and post to a general Slack channel. 6. If false: Route to specific Slack channels (#billing-support, #tech-support, etc.) including the AI draft and internal notes. Deliverables: A valid n8n workflow JSON file and a setup guide. Quality Check: Run a dry test with 3 sample tickets (one simple, one complex, one security-related). Verify correct routing and email formatting. Do not activate the workflow for live traffic until I confirm the test results.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.