TL;DR
Before asking where AI could help an IT department, I would look at the work it is already doing. Ticket data shows what repeats, how much time it consumes and where automation could actually matter. Sometimes the answer is AI. Sometimes it is PowerShell, an API or a better process.
What the tickets revealed
We once analysed 35,233 IT tickets.
The interesting number was not the ticket count. It was what was hiding inside it: roughly 4,623.8 hours of work, worth about €393,000 a year under the assumptions we used.
Most of it did not look dramatic.
Password resets. Access rights. Mailboxes. Software. Printers. User administration. Monitoring. Small tasks that take five, ten or twenty minutes and do not feel important on their own.
Do them thousands of times and the picture changes.
That is why I like ticket analysis as a starting point for automation. The data already exists. It shows what the IT department is actually spending time on instead of what people remember from the last workshop.
You can cluster recurring work, estimate handling time, look at volume and then ask a very practical question: which of these tasks should still require a person?
That produces a much better discussion than starting with a blank whiteboard and asking everyone for AI ideas.
The right tool for the work
One process might need an LLM. Another is completely deterministic and should be five API calls. Another might be solved with PowerShell. Sometimes the process itself is the problem and should simply be changed.
I genuinely do not care which technology wins.
If a script removes the work more reliably than an agent, use the script.
Knowledge beyond the documentation
The other thing ticket data exposes is how much operational knowledge sits in people’s heads. An experienced admin reads a request and knows what the user probably means, which system is involved, what to check and where the exceptions are. Process documentation often captures only part of that. Thousands of historical tickets contain much more of the real workflow.
That becomes important once you try to automate it.
Remove the work, not just the queue
A ticket system can route a task perfectly, assign it, escalate it and measure the SLA. Somebody may still have to do the task.
That is the distinction I care about.
The goal is not a better organised queue. The goal is to find the recurring work that can disappear from the queue entirely.
And for that, a few tens of thousands of boring tickets are often a very good place to start.
