A mid-size SaaS company was drowning in support tickets: every one was read, tagged, and routed by hand before an engineer ever saw it. We built an LLM-powered triage agent that classifies, prioritizes, and routes tickets — with a human approval loop for anything uncertain.
Around 400 tickets a day, three time zones, and a two-hour median time just to get a ticket in front of the right person. Misrouted tickets bounced between teams for days.
Median time-to-routing dropped from two hours to under a minute, with routing accuracy above the human baseline on the evaluation set. The team now spends its time answering tickets, not sorting them.