When AI projects in IT service management fall short, the most common response is to blame the technology. The tool didn’t integrate well, the model wasn’t trained on the right data, or the vendor oversold the capability. Sometimes that’s true. More often, it isn’t. More often, the technology has landed on an organizational foundation that wasn’t ready for it, and the failure was baked in before a single workflow was automated.
This was one of the clearest threads running through SPARK26, the annual SDI conference held in March 2026, where speakers and practitioners worked through some of the harder questions the industry is grappling with on AI, operational maturity, and the changing shape of service management. Across several sessions, the same argument surfaced from different angles: AI readiness in ITSM is primarily a human and organizational problem, and treating it as a technology-selection problem leaves organizations disappointed.
The rocket fuel problem
Matt Beran, a well-known voice in the ITSM industry, spoke at SPARK26 on experimentation and organizational learning, and the IT teams seeing real returns from AI investment came up directly. They’re not distinguished by the sophistication of their AI approach. They’re distinguished by the work they did before AI entered the picture: documentation, process discipline, operational structure. “AI multiplies what you already have,” Matt said. “It’s rocket fuel if you’ve got great processes. But if there’s anything bad in there, that bad gets that much worse because it’s got rocket fuel on it.”
That framing, AI as amplifier rather than transformer, matters more than most organizations are ready to admit. It means the quality of your AI output is a direct reflection of the quality of your underlying data, your process documentation, your knowledge base, and your CMDB. If those are in poor shape, AI doesn’t fix them. It accelerates them.
A lot of what gets called AI readiness work is really just operational maturity work that should have happened anyway. Cleaning up resolution notes so they contain context rather than a single word like “fixed.” Ensuring categorization is consistent across the organization, so that what one analyst logs as “network issue” isn’t logged by another as “connectivity drop.” Retiring knowledge articles that are out of date or structured as unwieldy PDFs rather than content AI can understand. None of this is glamorous, and none of it is new. But organizations that skip it and expect AI to paper over the gaps are the ones most likely to report that AI hasn’t delivered, and to reach for a different tool rather than look at the foundation underneath.
The design problem underneath the data problem
A related but distinct argument came from Mark Boyer, who spoke at SPARK26 on the future shape of service management. Mark’s contention is that many service desks aren’t broken. They’re compensating. They’re compensating for services that were never designed to be easy to live with. And what AI does, in that context, is automate the compensation rather than address the design failure underneath it.
Picture a payroll manager on the last working day of the month. They open their system, and something fails. In the current model, the service desk asks that person to translate their dread into a category selection. A dropdown. The ticket gets logged as “keyboard not working.” The team closes it within the SLA. Everyone looks efficient. The user is no less anxious, because it was never a keyboard issue. It was the risk that 300 people might not get paid that day, and the system required the user to do the translation work, to guess the right category, to feed the process rather than be served by it.
The design question that rarely gets asked is: why did that friction exist in the first place? AI, applied well, can remove it: capturing intent in plain language, pulling context automatically, running diagnostics in the background, and keeping the user informed throughout. But AI applied to a poorly designed front door still produces a poorly designed experience, just faster. “Brilliance in the desk is often evidence of failure in the service,” Mark argued at SPARK26. If your service desk team is exceptional, part of the leadership question you should be asking is what they’re compensating for, and whether AI is being used to fix the underlying problem or to make the workaround more efficient.
This connects back to Matt’s point about foundations. Organizations that deploy AI on top of under-documented processes and poorly structured data aren’t just wasting money on a tool. They’re scaling the problem. Contacts that reach human agents become harder because the easy ones were routed away, and the friction remains. The system appears to be working because tickets are closing, but the user experience is getting worse.
The human paradox
Barclay Rae, founder of Barclay Rae Consulting and one of the more experienced observers of the service desk industry, identified a parallel dynamic at SPARK26 that compounds both of the above. As AI absorbs routine interactions, the interactions that reach a human agent become progressively more challenging: edge cases, emotionally charged conversations, trust-repair moments, and work that requires judgment. The paradox is that many organizations are treating AI adoption as a reason to reduce investment in human skills, at precisely the moment when those skills are becoming more critical.
Barclay’s point is that organizations may have fewer interactions than they used to, but that doesn’t mean the bar has lowered. The expectation for remaining human contact is higher than it’s ever been, because users who reach a human agent will already have tried the automated options. They’re coming with more complex problems, more frustration, and a lower tolerance for responses that don’t meet them where they are.
This changes what good looks like in a service desk role. The value of an analyst is no longer measured primarily in ticket throughput. It’s measured in judgment, empathy, and the ability to work at three levels simultaneously: the technical issue, the business context, and the human reality of the person reporting it. That’s not a skill AI can replicate, and it’s a skill that gets undervalued when organizations look at headcount and see automation as a straight substitution.
What readiness looks like
The organizations that will get meaningful returns from AI investment aren’t the ones that deployed fastest. They’re the ones that were honest about the state of what they already had, did the work to bring it up to standard, and then used AI to amplify something that was already functioning well.
That’s a harder and less exciting story than the one most organizations want to hear. It requires acknowledging that your CMDB might not be trustworthy, that your knowledge base probably isn’t structured for machine parsing, and that your categorization taxonomy was designed for human routing rather than pattern recognition. It requires investing in the unglamorous groundwork before expecting the technology to deliver.
It also requires taking the human capability question seriously. As automation handles more of the routine, the people remaining in service management roles need to be better equipped, not just fewer in number: more skilled at judgment, more capable of genuine empathy, more able to work across the technical, business, and human dimensions of a problem at the same time. Investing in that capability isn’t in tension with AI adoption. It’s a prerequisite for getting anything meaningful out of it.
SymphonyAI recognizes this. Getting the most from AI in service management depends on the quality of what’s underneath it and closing the gap between where organizations think they are and where they actually are is where the real work begins. AI deployed on a weak foundation will never succeed. That’s not a technological problem. It’s a human one. Read more here.