My First n8n workflow · AI Automation Agency Hub

My First n8n workflow

Just finished my first real n8n build — an AI Ticket Triage Assistant for ITSM, using n8n + Gemini.

Quick context: I work as an ITSM Analyst day-to-day, so I built something close to my actual domain — support tickets come in as messy free text, and normally a human has to read, judge urgency, and route each one manually.

v1 pipeline (kept deliberately simple, no agents/chatbots):

Raw ticket text goes in
Gemini extracts structured data — category, priority, impacted service, suggested team, summary
Deterministic rules (not AI) make the final routing call — e.g. High priority always goes to Major Incident Team first
Every decision logs to Google Sheets, AI's suggestion vs. the system's final call, side by side

Design choice I stand behind: AI suggests, business rules decide, everything's auditable. Didn't want a black box, especially thinking about how this would actually need to work in a real ITSM environment.

Bugs I hit (sharing since debugging seems to be most of the actual skill):

Deprecated Gemini model throwing silent 404s
Google Sheets OAuth needed both Sheets AND Drive APIs enabled
A wiring mistake silently misrouted tickets — only caught because I tested varied fake data instead of trusting the first clean run
Gemini initially over-weighted urgency language ("asap") over actual business impact — fixed by separating requester tone from objective impact in the prompt Tested against 10 varied fake tickets, all routed correctly.

Planning v2 next: multi-issue ticket detection, since real tickets are rarely this clean.

Would love feedback from anyone who's built something similar — especially on the human-in-the-loop pattern (AI suggests / rules decide / log everything). Is that overkill for v1, or the right instinct to keep from the start?

13 comments

My First n8n workflow