AI-powered ticket triage agent with SLA monitoring and escalation, built for GrowthLab, a performance-marketing agency.

Context

GrowthLab is a performance-marketing agency running ad campaigns (Google/Facebook Ads) for e-commerce clients. Customer inquiries arrived by email with no automated handling at all: account managers read every message themselves, judged category and priority "by eye," and tracked SLA deadlines by hand.

Scale (approx.): 20–30 client emails per day handled by a small team of account managers.

Before: Gmail and manual tracking — no ticket log, no automatic priorities and no escalation rules.

Problem

— Incoming customer inquiries in Gmail had no automatic classification by category or priority

— Nobody tracked SLA violations, so overdue tickets could sit unnoticed for hours

— No clear escalation mechanism existed for when a ticket owner failed to respond in time

— Account managers spent several minutes per email manually judging urgency

— Leadership had no visibility into which tickets were breaching SLA or being repeatedly delayed

Solution

I built an n8n workflow with two parallel branches: ticket classification and SLA monitoring with escalation.

The system: — automatically picks up every incoming Gmail ticket — classifies it by category, priority and customer sentiment using an AI agent (Claude Sonnet 4.5) with a Chain-of-Thought prompt — applies a VIP rule so clients with a budget ≥$5000/month are never under-prioritized — logs every ticket to Google Sheets and notifies the owner in Slack — checks SLA deadlines every 5 minutes and escalates overdue tickets across 3 levels (Slack reminder → Slack team-lead alert → director email) — alerts the team in Slack if an escalation email itself fails to deliver

Process

1. Discovery — analyzed which types of client emails the agency receives, how urgent each one is and what happens when an answer is late
2. Solution design — defined categories, P1–P4 priorities with response deadlines (SLA), a VIP rule for large clients and a 3-level escalation chain
3. Build — two parallel n8n workflows: AI classification of every new email (logged to Google Sheets + Slack alert to the owner) and SLA monitoring every 5 minutes with escalation
4. Testing — ran 10 test tickets across all categories and priorities, including a VIP client with a repeated problem; found and fixed 2 bugs in the escalation logic
5. Launch — connected a real escalation mailbox and verified all 3 escalation levels end-to-end, including a real email to the director
6. Optimization — added alerts in case an escalation email itself fails, and status updates so the same ticket is never escalated twice