AI is hitting its ROI targets in IT service management, yet 52% of teams say it increased their workload and only 7% say costs matched the plan. The problem isn't adopting AI, it's operationalizing it.

[For more news, click here] 

Artificial intelligence was meant to be simple to use. For the IT teams that run corporate help desks and incident queues, humans were meant to get their time back from repetitive tasks and allocate their expertise elsewhere. Two years into this experiment, tasks have been handed off. The extra time, however, has not been handed back.

In the 2026 State of ITSM Report from SolarWinds, the Austin-based IT management software company, out of 800 IT professionals, 84 per cent have stated that AI has met or exceeded their return-on-investment expectations. One line further, 52 per cent of the respondents say their overall workload has actually increased since adopting AI. 7 per cent said the cost of adoption matched what they planned for.

After roughly 16 months of running AI in their environments, in other words, most teams are not yet collecting a lighter load. Instead, the spent time managing the machine.

The Productivity and Cost Paradox

The productivity gains are real and measurable. Respondents said AI saves an average of 3.2 hours a week on detecting and flagging issues, 3.0 hours on end-user requests, and 2.9 hours on ticket triage. The catch is where those recovered hours go. Nearly half of respondents, 48 per cent, now spend meaningful time managing and maintaining AI tools and integrations. Another 47 per cent spend it reviewing and validating what the AI produces, and 37 per cent on training and fine-tuning models. The work AI removed has been replaced by a new category of work that barely existed at this scale a few years ago.

The costs follow the same shape. The expenses that most often caught teams off guard were staff training at 48 per cent, data quality and cleanup at 47 per cent, and ongoing tuning and maintenance at 45 per cent, none of which are one-time setup items. They are recurring parts of the operating model. More than four in five respondents, 83 per cent, now spend three or more hours a week simply keeping their AI systems running reliably.

That gap between sticker price and true cost is not unique to SolarWinds' customers, Gartner has predicted that more than 40 per cent of agentic AI projects will be canceled by the end of 2027, citing escalating costs, unclear business value, and inadequate risk controls rather than any failure of the underlying technology. The through-line is consistent: the model is the cheap part, and the operating costs around it is where the budget goes.

A Pattern Bigger than the Service Desk

The service desk is a useful place to watch this unfold, because ITSM is one of the more mature corners of enterprise AI adoption. The paradox it exposes, however, appears wherever AI meets real work. A 2024 study from the Upwork Research Institute found that 77 per cent of employees using AI said the tools had actually added to their workload, even as 96 per cent of executives expected AI to raise productivity. And MIT researchers, in a widely cited 2025 report, found that roughly 95 percent of enterprise generative AI pilots produced no measurable impact on profit and loss, with integration, not model quality, identified as the reason.

Read together, these findings describe one problem. Adopting AI has become easy while operationalizing it, so that it removes work rather than relocating it, has not. That distinction is what the SolarWinds data keeps circling back to.

“We're at an inflection point in IT service management. AI adoption is no longer the hard part — the hard part is building the organizational discipline to make AI actually deliver,” said Brad McGinity, GM of ITSM at SolarWinds. “The teams that get this right aren't just running a faster service desk; they're running a fundamentally different operation. At SolarWinds, our job is to make that transition as straightforward as possible — giving customers the platform, the data foundation, and the governance they need to move from AI activity to real AI payoff.”

Reactive, not yet Proactive

The report also pinpoints where the untapped value sits by asking where AI has had the greatest impact across the incident lifecycle. Teams named identifying issues before they hit users (31 per cent) and prioritizing and routing issues (23 per cent). Both are fundamentally reactive, responses to problems that have already surfaced. Only 19 per cent cited preventing issues before they occur, which is where the real leverage lies.

Moving AI upstream into prevention requires clean data, connected systems, and organizational discipline that most teams have not yet built. 85 per cent of organizations said their AI budget for ITSM rose year over year, 36 per cent of them significantly, and agentic workflows, the most proactive category of AI, drew the highest expected investment growth of any area in the survey.

Measuring the Right Thing

One finding suggests why so many teams feel busier rather than lighter. Only 21 per cent of respondents measure AI in terms of outcomes or experience rather than raw activity, and the teams that track activity instead of outcomes were 2.4 times more likely to report that their workload had increased since adopting AI. In this case, counting the tickets AI touches may show motion, measuring whether they produced a better resolution shows whether it counts.

The other recurring theme is that data, not algorithms, is the binding constraint. Data quality was the single most common reason AI failed to deliver expected value, which reframes data cleanup as part of the AI strategy rather than a separate project to be deferred. 82 per cent of organizations said they offer formal AI training and structured change management, and 66 per cent of respondents said their bonuses and performance reviews are now tied to AI efficiency gains.

The Middle East's Compressed Timeline

Nowhere are the stakes per month higher than in the Gulf, where national AI strategies and heavy enterprise investment are pushing adoption faster than the global average. PwC has projected that AI could add $320 billion to the Middle East economy by 2030, equivalent to 11 per cent of the region's GDP (PwC Middle East), and that scale of ambition compresses the timeline for getting AI right.

“The pace and scale of AI investment in the region means the gap this report identifies, between adoption and real operational payoff, is being compressed into a much shorter window, and the resulting business impact is amplified. The takeaway then is that AI adoption alone is never a guarantee of success. Without the same governance and data discipline the report calls out, speed just gets you to the workload problem sooner,” said Abdul Rehman Tariq Butt, Regional Director for the Middle East at SolarWinds.

It is a fitting place to end, because it captures what the whole report is really arguing. The organizations pulling ahead with AI are not the ones that adopted first or spent the most. They are the ones doing the unglamorous work of fixing their data, connecting their systems, and measuring outcomes rather than activity.

Related Articles

How Can Gulf Enterprises Measure Their AI Security Readiness? SANS Institute Has a Framework

Cequence Security Launches Agent Personas to Automate AI Agent Governance Across the Enterprise

Amber Lands €7M to Turn Enterprise Knowledge into Autonomous AI