AI rollouts are everywhere right now. Turning the time and money behind them into meaningful results is proving much harder.
Many organisations are in the same position. Teams rush to launch a new tool without a proper adoption plan, clear governance, or a way to track results (and sometimes all three are missing). Inevitably the tool then goes on to provide little value at best, and to cause serious risk for the business at worst.
MIT NANDA’S research puts the scale of it in perspective. Its State of AI in Business report found that “95% of organisations are getting zero return” on their AI investment. That’s despite an estimated $30 to 40 billion enterprise spend on GenAI.
Other projects stall before they even get that far. Gartner expects “over 40% of agentic AI projects will be cancelled by the end of 2027”, with rising costs, weak risk controls, and unclear evidence of value as the main causes.
There’s huge potential in AI when it’s deployed well, though. From choosing the wrong use case to overlooking governance, here are seven common mistakes organisations make when rolling out AI, along with how you can avoid them.
👇 Register below for our latest webinar on AI governance, including a chance to ask your own questions
1. Focusing on the tool before the problem
It’s easy to get swept up in what a new AI tool can do. That said, an impressive list of features means little if none of it addresses a problem your organisation needs to solve.
Start by working backwards. What are you trying to improve, and what’s getting in the way today? For an ITSM team, that might be analysts losing time to manual ticket categorisation, or users who can’t find the right answer in self-service. Once the problem is clear, you can decide whether AI is the right fix and what capabilities you need.
This also gives you something concrete to measure later. If the goal was to drop categorisation time, for example, you can check whether this has reduced, instead of relying on usage figures alone.
👉 More from GEL: What is AI governance?
2. Feeding AI scattered and poor data
AI can only give reliable answers about your environment from what it can access. If that information is scattered across systems, buried in old tickets, or sitting in an outdated knowledge base, its answers will only ever give you part of the picture.
For ITSM teams, that can cause problems quickly. An AI assistant might read your knowledge base, but miss the fix sitting in a previous incident. Or it could confidently surface guidance that’s several versions out of date.
Before rollout, map out what information your AI will rely on and where it lives. Make sure it can access everything it needs to – within your governance guardrails, of course.
3. Assuming AI will fix underlying problems
AI can certainly speed up processes, but that doesn’t mean those processes are ready to be automated. If a workflow has too many unnecessary hand-offs or relies on unclear decision-making, adding AI can carry those same problems forward.
Take incident management. If tickets regularly bounce between teams because nobody’s sure who owns what, automating the routing won’t address the underlying issue. You could simply end up moving tickets through a flawed process faster.
Before introducing AI, look at the process itself. Work out what could be simplified and where human judgement still matters. You’ll then have a much clearer idea of where AI can add value.
👉 Read next: ITIL® 4 vs ITIL (Version 5): what’s changed?
4. Giving people AI without showing them how to use it
Skills gaps are already a problem for employers, and AI is adding to the pressure. BusinessLDN research found that only 50% of London businesses believe their workforce has the skills and capabilities they need, down from 63% the year before. And amongst the firms already using AI, 85% said it had changed the skills they need.
That’s why an upskilling strategy should be built into your implementation plan from the start. An AI tool only turns into value when the people using it know how to get the best out of it. Show them how to use it within their role, including how to write clear prompts and give it the right context. The key is clarity on what AI is appropriate for and what information is safe to put into it. A service desk agent, for example, should know whether pasting a customer’s details into a prompt is allowed.
Consider the user experience as part of this too. If people struggle to fit the tool into their existing workflow, adoption can quickly drop once the initial enthusiasm wears off.
5. Overlooking what happens to AI outputs
AI policies often focus heavily on what people put into a tool, particularly sensitive or confidential data. What happens to the output deserves just as much attention.
AI can produce information that sounds convincing but is inaccurate or based on the wrong context. In an ITSM setting, that becomes an issue if an analyst uses an AI-generated resolution without checking it, or if incorrect guidance makes its way into the knowledge base and is then surfaced again later.
The level of oversight should reflect the consequences of getting something wrong. Your governance needs to make clear when an output should be checked and who can approve higher-risk uses.
👉 More from GEL: The shifting role of IT in 2026 and beyond
6. Not measuring the impact
Most leaders can’t say with total certainty what their AI is delivering. Research from Catalyst across the UK, France, and Germany found that 95% can’t quantify the impact or value as their business simply isn’t tracking it.
Measuring AI’s impact is tricky but start with the problem you set out to solve. For example, if the goal in our first point was to cut the time analysts spend categorising tickets, compare categorisation times before and after rollout. Check accuracy too, so speed isn’t costing you misrouted tickets.
Cost belongs in this conversation too. Licences and ongoing usage add up as a pilot scales, and measuring the value delivered against that spend helps you decide whether to keep investing or put the budget somewhere else.
7. Lack of clear ownership
Who owns your AI? When a tool cuts across teams and systems, the answer isn’t always obvious. IT might manage the technology, another team might own the process it supports, while employees across the business use it day to day.
That makes clear ownership essential. Every AI deployment needs someone accountable for it throughout its lifecycle, with the authority to make decisions about how it’s used and when intervention is needed.
An AI agent that resolves or routes incidents on its own is a good example. There should be a named owner responsible for how it performs, who’s empowered to step in if it starts making poor decisions. Without that clarity, problems can fall between teams and decisions can be delayed when someone needs to act quickly.
👉 Read next: The 7 guiding principles of ITIL® (Version 5)
The AI governance gap: ask your questions to an ITIL expert
When an AI mistake happens inside your business, who is accountable? Most organisations can’t answer that with confidence – and that’s a big risk.
Join Roman Zhuravlev, Senior Architect at ITIL®, for our live webinar on Thursday 29th October to explore how to deploy AI confidently and responsibly, while still delivering business value.
📅 Date: Thursday 29 October
🕐 Time: 2.30pm (GMT)
🎥 Can’t attend live? Register and we’ll send you the recording.
