IT teams are getting measurable returns from AI in service management, but most are still not there yet. TeamDynamix surveyed organisations already using AI in ITSM and found 82% reporting ticket deflection, 71% faster resolution times, and 76% improved customer satisfaction. A separate Atomicwork study with ITSM.tools puts the share of IT professionals reporting positive ROI at 82%, up from 24% a year earlier.
Both reports come from vendors with a stake in the answer, but the consistency across them is notable. Something is working for the organisations that have got there. The question is what they’re doing that most haven’t.
What AI in ITSM Is Actually Being Used For
One finding from the TeamDynamix research stands out: the top use case isn’t virtual support agents. It’s knowledge base gap-finding and content generation, cited by 88% of participants, ahead of ticket deflection and intelligent triage.
That ordering is significant. Knowledge management is the foundation that determines whether the more visible use cases actually work. A virtual agent that can’t surface the right article fails at first contact. Triage logic routing to the wrong category creates tickets rather than closing them. Teams that use AI to improve their knowledge base first are building what the rest of the stack depends on. Freshworks made exactly this argument in April, framing its Freshservice ITAM update around messy infrastructure data as the thing most likely to undermine AI deployed on top of it.
Atomicwork’s research also found that organisations where IT leadership drives AI investment, rather than fragmented departmental experimentation, show higher trust and more sustained outcomes. In 2026, 54% of AI investments were initiated by IT leadership, up sharply from previous years.
The Gap Between Policy and Practice
Auvik’s 2026 IT Trends Report found three-quarters of IT leaders believe their organisation has an AI policy. Fewer than half of help desk staff say the same. Auvik CEO Doug Murray says:
“When three-quarters of IT leaders believe they have an AI policy but fewer than half of help desk staff say the same, that’s an implementation problem versus a policy problem. Until governance is understood at every level of the organisation, AI risks becoming just another source of shadow IT rather than a solution to it.”
That disconnect explains a lot of stalled deployments. We covered the same problem in May, looking at why most enterprises aren’t ready for the agentic ITSM tools Ivanti and ServiceNow are already shipping: stale CMDBs, missing API governance, and no visibility into what agents are actually doing. Those aren’t consequences of moving too fast. They’re the reasons AI doesn’t perform when you do move.
Why Architecture Matters More Than the AI Itself
The TeamDynamix report identifies integration complexity as the second major barrier after data readiness. Organisations running AI outside their core ITSM workflows report friction, data silos, and limited returns. OpenText’s research on ITSM AI adoption found a specific fault line: organisations using public LLMs without platform integration reported measurably worse outcomes than those running AI natively inside their ITSM environment.
That context makes the ServiceNow and Accenture Forward Deployed Engineering programme worth reading alongside this data. The programme places embedded engineering teams inside customer environments to close exactly that integration gap. Accenture’s Ram Ramalingam says:
“The question our clients ask is not whether to invest in AI. It’s how to make it work at enterprise scale.”
Accenture’s Pulse of Change research finds only 32% of leaders report sustained, enterprise-wide AI impact. Vendors from ServiceNow to Ivanti and Workday have all made the same call in recent product releases: AI built into the platform outperforms AI added onto it. The outcome data is starting to back that up.
What Buyers Should Take From This
Eighty-seven percent of organisations are either already using AI in ITSM or plan to within 24 months, per TeamDynamix. The teams hitting 50%-plus deflection rates aren’t running more sophisticated AI than everyone else. They’re running AI that can see the tickets, the assets, and the knowledge base in one place. As Cisco’s Amit Barave told UC Today this week, the conversation has already moved on from whether to adopt. The practical question is whether the foundations are solid enough to get value when you do.
