Fix your ticket categories before you ask AI to triage anything

Most service desks that try to automate ticket triage do it in the wrong order: point a model at incoming tickets and an existing category list, and get fast, confident categorisation that's exactly as inconsistent as the category list it was given - "Access Issue", "Login Problem" and "Can't Get In" all still exist as separate categories, just applied more quickly and by something that won't push back on the overlap the way a human triager occasionally does out of irritation.

Before automating triage, ask the model to audit the category list itself - paste every category currently in use, along with a sample of tickets filed under each, and ask it to flag categories that overlap, categories with almost nothing filed under them, and tickets that clearly don't fit the category they landed in. Fix the list based on what comes back, then automate triage against the cleaned-up version.

Skipping this step doesn't just produce messy reporting - it actively hides trend data you'd otherwise catch. Twelve tickets a month about the same recurring fault, split across three overlapping category names, never crosses whatever threshold triggers "this needs a permanent fix" in your reporting. Under one consistent category, it does.