AI Ticket Routing + SLA + Escalation — Automated Support Triage & Accountability

How a building-materials retailer replaced manual inbox triage with an AI dispatcher that classifies, routes, and escalates every support ticket on its own

🏢 Context

БудМак — a retail + wholesale building-materials store serving both individual customers and B2B clients (foremen, developers) with orders ranging from single deliveries to large construction-site supply contracts. Customer support came in almost entirely through one shared Gmail inbox.

Every incoming email — a damaged-delivery complaint, a request for a wholesale quote, a payment error, a thank-you note — landed in the same place and was triaged by whichever manager happened to check the inbox first.

❌ Problem

Support had no structure, no SLA, and no accountability layer:

• No consistent triage — priority depended on which manager opened the email first, not on actual urgency (a bricklaying crew standing idle waiting for a delayed concrete truck got the same treatment as a “thanks!” reply)
• No SLA tracking — nobody knew a ticket was overdue until the customer complained again, angrier the second time
• No escalation path — if an owner missed a ticket, it simply sat there; there was no automatic handoff to a team lead or director
• Inbox noise — newsletters, “unsubscribe” spam, and no-reply notifications mixed in with real client emails, wasting manager attention
• No visibility into risk — legally sensitive complaints (threats of court action, tax authority mentions) or payment-gateway failures where money was charged but no order existed could sit unnoticed for hours

💡 Solution

Two connected n8n workflows that turn a shared inbox into a self-monitoring support system:

1. AI Ticket Classifier polls Gmail, filters out non-client noise, and runs every real client email through a chain-of-thought GPT-4o-mini agent that assigns category, priority, owner, and SLA deadline — then logs the ticket and pings the owner in Slack, all within minutes of arrival
2. SLA Watcher runs independently on its own schedule, checks every open ticket against its deadline, and escalates through three levels — owner reminder → team lead → director email — automatically, with no duplicate pings once a level has already fired

The key design principle: classification and enforcement are two separate, independently-scheduled workflows. The AI never has to “remember” to check deadlines — a dedicated watcher does that continuously, driven purely by data already sitting in the sheet.