AI Ticket Triage & Insights

The team was receiving more tickets each week than anyone could sort by hand. Triage happened when someone had a spare moment, which meant urgent issues sat behind routine ones and nobody could tell which was which without opening every ticket.

Categorisation existed on paper. In practice each agent applied it slightly differently, so the reporting built on top of those categories described nothing anyone trusted.

We trained a classification model on thousands of historical tickets, learning the categories, sub-categories and priorities the team had already assigned. Their own past decisions became the training signal, which meant the system learned how this desk actually works rather than how a generic taxonomy says a desk should work.

The model went into production in front of the live queue. Every incoming ticket now gets classified and routed the moment it arrives, with no manual sorting step. Each ticket also carries a summary written by an LLM, so an agent has the context before opening it.

A second automation runs weekly. It takes that week's ticket summaries and rolls them into a single digest, surfacing recurring patterns and turning them into concrete action items for the engineering team. Finding those patterns used to mean hours of reading.

Triage stopped being a job someone does and became something that happens. The weekly digest gave the engineering team a prioritised list drawn from what customers actually reported, rather than from whoever complained loudest that week.

The client's name is withheld at their request. The pattern transfers to any business with a queue of inbound text: enquiries, claims, applications, maintenance requests.