We replaced a 12-person support triage queue with one GPT-4 workflow

A SaaS client's support team spent the first 20 minutes of every ticket just figuring out where it should go — billing, technical, or product feedback — before anyone could actually help the customer. With 300+ tickets a day across a 12-person team, that triage overhead alone ate the equivalent of two full-time roles.

We built a single AI triage workflow that reads, classifies, and routes every incoming ticket before a human ever sees it. First-response time dropped from 6 hours to under 2 minutes.

The real cost of manual triage

Triage isn't just slow — it's inconsistent. Different agents categorized similar tickets differently, which meant reporting on ticket volume by category was unreliable, and escalation rules (like "billing disputes go straight to a senior agent") were applied inconsistently depending on who read the ticket first.

The architecture

Every new ticket triggers a webhook into a Make scenario. The ticket body and subject are sent to GPT-4 with a classification prompt, along with a short summary of the customer's account tier and recent activity pulled from the CRM — this context noticeably improves classification accuracy over the raw text alone.

Classification that actually holds up

The model returns a structured category, urgency score, and a one-line summary for the agent — not a full response draft. We deliberately kept AI out of writing customer replies for this client; the goal was routing speed, not response automation, since their support quality bar for tone was very specific.

Routing and escalation

Based on the urgency score and category, tickets are auto-assigned to the correct queue and, for anything flagged "billing dispute" or "churn risk," immediately escalated to a senior agent with a Slack ping — no human has to notice the pattern first.

The results

• First-response time dropped from ~6 hours to under 2 minutes
• Misrouted tickets dropped by 91% in the first month
• Senior agents now see churn-risk tickets within minutes instead of hours
• Reporting on ticket categories became reliable for the first time

"We didn't replace our support team — we gave back the 20 minutes per ticket they used to lose just figuring out where it belonged."

If your team is drowning in triage before the real work even starts, this is one of the highest-leverage automations we build. Book a discovery call to talk through your setup.