Which IT support processes are a good fit for AI workers

The question is rarely whether AI helps in IT support, but where to start sensibly. In our own operations we have measured that more than half of support hours sit in recurring, automatable tickets. That is exactly where the first lever is, not in the rare special case.

Good starting points

• 1st-level requests that frequently appear in a similar form.
• Recurring 2nd-level routines with clear steps.
• Onboarding steps for new employees.
• Standard configurations that run rule-based.
• Handovers and documentation that often get left behind in daily work.

These processes have three things in common that we see again and again: they are recurring, rule-based, and clearly scoped. And they can be given a clear escalation to humans in case a case falls outside the frame after all.

What does not (yet) fit

• Processes with unclear responsibility.
• Cases with high uncertainty or a lot of discretion.
• Workflows without defined approval rules.

We do not automate such processes from day one. Often it is still worth describing them cleanly first. Surprisingly often it turns out that part of them is in fact well scoped and works as a starting point.

How we scope a process

We start with the question of which steps run manually and always the same today. From this comes a clearly defined process with defined input, fixed steps, clear boundaries, and an escalation. Better a small, cleanly controlled process than a large one that in the end no one can review.

What this means for nordnung.ai

For nordnung.ai, such use cases are the concrete entry point. They are suited to trial a controlled AI worker with clear approvals and boundaries, instead of starting with a fuzzy large project. That is how we started ourselves.