Artificial intelligence (AI) is a force multiplier. Give it mature, well-structured processes (let’s call it “operational maturity”), and it will scale them. Give it chaos (inconsistent categorization, fragmented knowledge, undocumented workflows), and it will scale that instead. Faster than any human could.
I’ve spent the better part of the last few years talking to IT people about AI. And almost every conversation follows the same arc: excitement about the potential, frustration with the results, and a quiet search for someone to blame when the deployment doesn’t deliver.
The AI vendor usually gets blamed. Sometimes the team. Occasionally, the budget.
But in most of the cases I’ve seen, the real issue isn’t any of those things. It’s something more uncomfortable: the operation wasn’t ready…
This isn’t a fringe view. Research from MIT Project NANDA found that while around 60% of task-specific AI tools get evaluated in enterprise environments, only about 5% reach production. Gartner has flagged that by 2027, half of AI projects in IT service desks may be abandoned due to cost, risk, or failure to deliver ROI.
Operations and operational maturity are the bottleneck.
The AI-as-a-Feature Trap in IT Service Management
Part of the issue is how AI has been sold to IT service desk teams. In most platforms, it shows up as a standalone capability – a button, a chatbot, an add-on. Service desk agents are expected to “use AI” separately from their daily work.
The result is predictable. Adoption is inconsistent. Outputs feel disconnected from real tickets. Trust remains low. And because the tool sits outside the workflow, it rarely gets the chance to improve.
But there’s a deeper issue underneath this. When organizations layer AI onto immature operations, they don’t just get low adoption; they get automated fragility.
Poor knowledge quality doesn’t get fixed by an AI recommendation engine; it gets distributed faster. Inconsistent ticket categorization doesn’t improve when you add auto-routing; it creates mis-routing at scale. The issues that were slow and visible become fast and invisible.
That’s the chaos amplification problem. And it’s why I think the IT service management (ITSM) industry conversation about AI maturity is asking the wrong question.
The Wrong Question: “How Can We Automate More?”
Most maturity models in ITSM (and most ITSM tool vendor roadmaps) frame progress as a march toward greater autonomy. The destination is an IT service desk where AI handles the volume, deflects the tickets, and resolves issues without human involvement.
I understand the appeal. The economics are obvious. But treating autonomy as the goal creates a dangerous incentive: to deploy automation before the foundation is ready, because the pressure to show results is real and the consequences of a fragile deployment take time to surface.
The better question is: What does our operation need to look like before AI can be trusted to act?
Answering this operational maturity question honestly is harder. It requires looking at knowledge quality, workflow consistency, ticket data hygiene, and ownership clarity; none of which are as exciting as a demo of a virtual agent resolving a password reset. But they’re what determine whether AI becomes a genuine multiplier or an expensive liability.
The AI Adoption Lifecycle: A Framework for Sustainable AI in ITSM
We’ve been working with this question for a while, and what we’ve arrived at is a framework we call the AI Adoption Lifecycle. Rather than describing operational maturity as levels (implying a single destination and a sequential climb), we think of it as layers of expanding intelligence, each one building the trust and data quality that the next requires.
Operational Maturity Layer 1 – Assisted Intelligence
At this layer, AI supports agents during daily work – but only when they explicitly ask for it. Ticket summarization, keyword generation, solution recommendations, response drafting. Nothing happens without the agent’s choice.
The point isn’t efficiency yet; it’s trust and data quality. Agents start to see that AI suggestions are grounded in real ticket history and approved knowledge. The data starts to improve because AI surfaces gaps. New service desk agents ramp faster. The foundation gets stronger.
Operational Maturity Layer 2 – Embedded Intelligence
This is the most important shift. AI stops looking at individual tickets and starts learning from operations over time. It identifies recurring issues, detects emerging incidents before they’re declared, flags sentiment shifts in ticket conversations, and surfaces knowledge gaps.
Critically, it can also extract knowledge from resolved tickets, turning the institutional memory trapped in closed tickets into structured, validated articles that improve future recommendations.
Teams that operate at this layer stop firefighting. They gain earlier visibility, fewer duplicate incidents, and better decision-making context. AI hasn’t taken over anything – it’s made humans more effective.
Operational Maturity Layer 3 – Governed Intelligence
Only here does AI interact autonomously with end-users, and only within explicitly defined governance boundaries. Routine, well-understood requests. Clear escalation paths. Validated knowledge built through the layers that came before. Automation at this stage is a natural consequence of operational maturity, not a shortcut to it.
The distinction matters: AI handles high-volume, low-risk interactions. Anything ambiguous, sensitive, or outside defined parameters routes to a human, with full context already assembled.
Governance is what makes expansion safe. By defining what knowledge is trusted and what actions are allowed, teams can extend AI’s reach without introducing chaos.
The Hidden Governance Gap in Enterprise AI Adoption
There’s a pattern I see repeatedly that illustrates the stakes. Organizations deploy AI tools, adoption is inconsistent, and agents start routing around the official tool, copying ticket details into external AI assistants to draft responses, and summarizing issues in consumer apps to save time.
This is rational behavior. The external tool is faster and feels more useful because it has fewer constraints. But it creates real issues: data-handling risks, compliance exposure, and an AI layer that management can’t see, govern, or improve.
MIT Project NANDA found that while only around 40% of organizations report purchasing official GenAI tools, over 90% of employees use personal AI tools at work.
The operational Maturity and governance gaps are already there; it’s just invisible. Building AI into the platforms where work already happens, and making that AI genuinely useful at each stage, is what closes it.
That’s why embedding matters as much as capability. An AI feature that agents actually use inside their workflow – even if it’s more constrained than a consumer tool – is worth more operationally than a sophisticated autonomous system that gets bypassed.
How to Assess Your Organization’s AI Readiness
If you’re an IT leader trying to make sense of where your organization actually stands in terms of operational maturity, here are the honest questions worth asking:
Is our knowledge base current, validated, and trusted?
If agents don’t trust the knowledge base for their own work, AI won’t be able to improve on it.
Are our tickets consistently categorized?
Inconsistent categorization is one of the most common reasons AI recommendations miss the mark.
Do agents have a reason to use AI in their daily workflow?
If the answer is no, start with assisted intelligence – build the habit before building the automation.
What are service desk agents doing with AI tools we didn’t deploy?
The answer will tell you more about your readiness gaps than any internal assessment.
Are we measuring adoption, not just deployment?
A feature being available is not the same as it being used.
Operational Maturity Comes Before AI Maturity
None of these questions are about technology. They’re about operations and operational maturity. And that’s the point. The IT service desks that are getting meaningful value from AI aren’t necessarily the ones with the most sophisticated tools –they’re the ones that did the operational maturity work first.
Operational Maturity FAQs
Operational maturity refers to how well an organization’s ITSM processes, knowledge, workflows, data quality, and governance are defined, maintained, and consistently executed. Mature operations provide the foundation needed for successful AI adoption.
AI depends on high-quality data, trusted knowledge, and consistent processes. Without operational maturity, AI can amplify existing inefficiencies, errors, and inconsistencies, leading to poor outcomes and lower return on investment (ROI).
AI can help identify process issues, but it cannot automatically fix poor operational practices. If workflows, ticket categorization, or knowledge management are inconsistent, AI may scale those issues rather than solve them.
Chaos amplification occurs when AI automates or accelerates flawed processes. For example, inaccurate knowledge articles, poor ticket categorization, or ineffective workflows can be propagated faster and at greater scale through AI-driven automation.
The AI Adoption Lifecycle is a framework that helps organizations implement AI progressively through three stages: Assisted Intelligence, Embedded Intelligence, and Governed Intelligence. Each stage builds the trust, governance, and data quality needed for more advanced AI capabilities.
Assisted Intelligence is the first stage of AI adoption, where AI supports service desk agents through capabilities such as ticket summarization, response drafting, solution recommendations, and knowledge suggestions. Human approval remains required for all actions.
Embedded Intelligence occurs when AI continuously analyzes operational data to identify trends, recurring issues, emerging incidents, knowledge gaps, and service risks. It helps teams make better decisions without taking autonomous actions.
Governed Intelligence is the stage where AI can autonomously handle specific low-risk and well-defined tasks within clearly established governance boundaries, while escalating exceptions and complex issues to human staff.
Organizations should introduce autonomous AI only after establishing trusted knowledge, consistent processes, reliable data, and clear governance controls. Autonomous AI should be the result of operational maturity rather than the starting point.
Organizations can evaluate AI readiness by examining the quality of their knowledge base, consistency of ticket categorization, AI adoption levels among agents, governance controls, and the extent of unofficial AI tool usage across the workforce.
Agents often turn to external AI tools when official solutions are difficult to access, poorly integrated into workflows, or provide limited value. This behavior can create governance, compliance, and data security risks.
Knowledge management is critical because AI recommendations, responses, and automations rely on trusted information sources. Poor-quality knowledge bases reduce AI accuracy and end-user trust.
Consistent ticket categorization improves AI training, routing, reporting, trend analysis, and recommendation accuracy. Inconsistent categorization makes it harder for AI systems to identify patterns and deliver reliable results.
Common causes include poor data quality, inadequate governance, fragmented knowledge, inconsistent processes, low user adoption, unrealistic expectations, and a lack of operational maturity.
Organizations should focus on adoption rather than deployment. Simply making AI available does not create value. Success comes from agents actively using AI within their workflows and achieving measurable operational improvements.
IT leaders can reduce governance risks by embedding AI into existing workflows, establishing clear policies, monitoring AI usage, validating knowledge sources, defining escalation rules, and providing approved alternatives to consumer AI tools.
No. AI maturity depends on operational maturity. Organizations cannot achieve sustainable AI success without first establishing strong operational foundations, governance practices, and reliable service management processes.
The first step is improving operational maturity by strengthening knowledge management, standardizing workflows, improving data quality, and ensuring service desk teams trust the information and processes they use every day.
Matt Beran
Matt Beran is an IT Industry Analyst and Expert who is currently the Product Marketing Specialist at InvGate. He also hosts the extremely popular Ticket Volume Podcast.
