A colleague made an observation recently that's been hard to shake: Perplexity already answers most IT support questions better than most software vendors' own support teams. So why bother building AI into your service desk at all?
Rob Garmaise, VP of Product at Info-Tech Research Group, had a direct answer. "Each client's data tells a very specific story," he said. "What infrastructure has been successful. What infrastructure has been less successful? Where are the challenges? Which users are causing the most problems? You can be that prescriptive."
General-purpose AI can tell you what good IT support looks like. Your own ticket data can tell you what's actually broken in your organization — and what to fix first.
The Accounting Audit Problem
Most organizations have years of ITSM ticket data sitting idle. They run periodic analyses, pull summary reports, and track SLA compliance. What almost nobody does is read every ticket.
"When you generally do a service desk ticket analysis, you don't read every ticket," Garmaise said. The analogy is a financial audit — you'd never manually review every transaction. But AI can.
And when you do, patterns emerge quickly. Which tickets are repetitive? What knowledge base articles are missing? Which monitoring alerts keep misfiring because nobody ever fixed the underlying diagnostic? "Really practical things," Garmaise said, "that work down the number of tickets over time."
That's the first of what he described as two basic vectors for AI-powered service desk analysis. The second is more interesting: combining unstructured data sets that humans rarely put together. Ticket data, satisfaction scores, application data plus cost data. Start correlating those, and you get answers to questions most IT teams have never been able to answer cleanly. What does it actually cost to service different user groups? Which groups are consistently hard to support — and why? Where are the highest-friction points in your application portfolio?
"Not rocket science," Garmaise acknowledged. "Just things that humans don't put together across so many data sets."
The Data Is Messier Than You'd Hope — But It Doesn't Matter
One practical concern with any AI-on-data initiative is data quality. Service desk tickets are not known for their literary precision. Garmaise was candid: "It's not all Shakespeare."
A lot of tickets are brief — "I need this" or "X, Y, Z happened." AI can augment those descriptions, inferring incident type, business impact, and cost of experience from context. But more importantly, augmentation isn't required to get useful output. "Just based on the descriptions we get, we can tell you here are the repetitive ones, here are the steps you need to take."
The cleanup burden, in other words, is lower than most IT leaders expect.
Where Organizations Actually Get Stuck
The bottleneck isn't data quality. It isn't tooling. It's a failure to go deep enough.
"They never get down to the data level," Garmaise said. "They stay at the process level. They have their processes working in various degrees of order, but none of them are reading the tickets and asking what is really happening in our operation."
Process-level thinking produces process-level answers. The specificity that drives real improvement — the kind that tells you which three ticket categories account for 40% of your volume, or which application is generating disproportionate support load — only emerges when you go to the data.
What This Means for Developers
Developer-generated tickets — environment issues, access requests, pipeline failures — get handled the same way as end-user tickets through AI analysis. Though Garmaise noted they're often more systemic. "A little more indicative of where you have alerts that are misfiring or not well managed."
But the more direct benefit for developers isn't about their own tickets. It's about everyone else's.
"What kills developers and engineers is when others are coming to their desk with questions — particularly low-value questions that take time," Garmaise said. "AI can knock all that out. Get rid of all the low-value foot traffic. Let them focus on what they do best."
That's not a small thing. Interrupt-driven support work is one of the most consistent productivity drains in engineering organizations. An AI layer that handles routine questions before they reach a developer's desk directly impacts output.
Two Years From Now
The near-term vision Garmaise described extends well beyond IT support. "We'll be running all corporate tickets through this process — HR, finance, customer support. All of it runs through the same process."
The metrics will shift accordingly. Less focus on SLA compliance, more on service uptime, robustness, and scalability. Less reactive measurement, more strategic view of how the organization is actually being served.
And the scope keeps expanding. "What's your application portfolio? How well optimized is it? Are you even investing in the right areas? Where are your gaps? How does that align with your required business capabilities?" Garmaise said. "There are all sorts of ways AI can help you stitch together the health of your organization. We're just scratching the surface."
The organizations that start going to the data level now — not the process level — are the ones that will have something to stitch together.
