Members of the itSMF UK Service Design and Transition Community of Practice (CoP) recently sat down for an open, roundtable discussion to tackle one of the most transformative, yet poorly guarded movements in modern ITSM; the practical adoption of AI. What emerged from the CoP practitioners, who work in a variety of public and private sector industries, was real-world battle scars, innovative custom tooling, and structural dependencies that AI must cross to be enterprise ready and deliver something of value. Here is what we discussed.
1. PERSONA IDENTITY FOR KNOWLEDGE DOCUMENTATION
Traditionally, knowledge articles and service design packages are written as single, sweeping documents meant for everyone. Predictably, half the audience finds them completely lacking in service details, while the other half gets lost in incomprehensible technical jargon or irrelevant acronyms.
“I took all of [the team’s] design knowledge, which is quite voluminous, and came up with some different personas, like ‘I know nothing about design and don’t care,’ ‘I am a project manager,’ or ‘I’m a CAB approver’… and then got AI to rewrite all of the content based on the point of view of the persona. You end up with a set of knowledge-based articles that are all the same content, but through a different filter… getting the language aimed specifically at the persona.”
To address this, members shared how they read through their knowledge and identified distinct personas that would help shape their stakeholders’ viewpoints and used a variety of Large Language Models (LLMs) to help re-write the content making it more relevant and suitable to the persona reader and the situation.
Since creating these specific personas within tools like Microsoft Copilot and custom code environments, the language used has seen organisations report a dramatic change in the value of their documents during critical timelines and a clear reduction in their MTTR (Mean Time to Resolution).
2. HIGH-LEVEL SYSTEM ARCHITECTURE BLUEPRINTS
Another highly practical use case originates from immature or under-resourced delivery environments where baseline technical blueprints simply do not exist. Service design architects are increasingly forced to manually cross-reference dozens of disparate documents from vendors just to get an accurate view on the components needed and how a service will be delivered.
To bypass this roadblock, practitioners are leveraging advanced visualization extensions, such as Atlassian’s Rovo AI within Confluence or tailored ChatGPT prompts, to sketch systemic boundaries. By feeding raw infrastructure components directly into a private or enterprise licensed LLM, designers are automating the creation of visually engaging high-level conceptual designs.
“Earlier, we used to have a lot of conversations with product and engineering teams to understand the high-level system landscape or system architecture… We use the Rovo AI within Confluence to set a very specific context… and it is doing a nice, visually engaging, high-level system landscape for us, which significantly reduces the amount of time we spend.”
Once these high-level architectural sketches are built by AI, they are taken directly into project forums to be validated by the relevant IT functions and human engineers.
3. SERVICE ACCEPTANCE CHECKLIST AND CAB ASSURANCE
One of the most immediate points of friction between Service Transition, Change Management, and broader Operations is the operational readiness review. Too often, technical project managers arrive at a Change Advisory Board (CAB) or operational gateway with critical gaps in their Service Acceptance Criteria (SAC) checklists, consuming massive amounts of human group time just to point out obvious non-compliance.
CoP practitioners are actively solving this overhead by implementing automated pre-check analysis scripts and custom interfaces that act as automated gatekeepers.
“Using this custom functionality that one of our developers built, we have some prompts that people can use when they’re creating changes to get an AI review… just helping people write good quality change, comparing it to standards that we’ve developed for what a ‘good change’ looks like… We’ve also got prompts for us to use to evaluate changes that are submitted… to give us some advice on whether it’s ready or not compared to our standards.”
By using AI to evaluate a draft Service Design Pack or change record against baseline SAC checklists, authors catch omissions in seconds.
4. DOWNSTREAM OPERATIONS CONTINUAL IMPROVEMENTS
The benefits of solid Service Design and Transition execution are traditionally felt downstream in operations, and this is exactly where AI dependencies tighten. The CoP noted that custom AI capabilities are scaling out across service desk environments to categorise, route, and analyse data into clean operational insights.
For instance, practitioners are using operational AI summaries to analyse highly complex incident commentaries and structural technical data into plain language known error records. This creates a direct feedback loop into the design phase for future releases.
“Fresh Service [Freddie AI] quite nicely summarises the incident for me… to have a look at what incidents have happened and what I can learn from it so that I can improve the processes and implement improved processes… helping us to capture the information in layman’s terms of a problem, an example is for known errors, we can save it into a knowledge article and make sure people understand what the problem is from a user’s perspective.”
Furthermore, practitioners highlighted how they utilise tailored tools like Microsoft “Cowork” skills to auto-generate standard client facing transition plans, standardised emails, and resource kits directly out of low-level designs at the simple click of a button.
5. CRITICAL GUARDRAILS: THE RULES OF AI ADOPTION
To successfully adopt these ways of working without compromising corporate integrity, the Community of Practice urges itSMF members to establish several foundational governance parameters:
- Implement a permanent human-in-the-loop: AI tools are built on a default preference to please the user, which frequently manifests as structural “hallucinations”. Managed service teams cited terrifying instances where standard bid text engines misapplied environmental data to major incident management scopes, creating entirely false work streams that would have cost thousands of pounds if left unvetted. Never export raw AI outputs directly to a client without thorough human validation.
- Establish a prompt library: To maximise prompt accuracy and guard against data drifts, standardising core structural boundaries is vital. Ensure prompts contain strict parameters:
“I say in my prompting, if you’re not sure of an answer, don’t just give me anything, just tell me that you’re not sure and that helps… It’s the kind of thing you need in a prompt library… a universal set of text that gives you the guard rails that apply to every single request: ‘Don’t make things up. If you don’t know, say you don’t know.'”
- Review and rationalise your process before automating: Throwing AI at a broken, bloated, or overly complicated legacy process merely automates and highlights structural waste. As a core design principle, lean optimisation must occur first.
“If you’re designing a process and you want to use orchestration, then you need to first review the process before you orchestrate it, because you’re going to orchestrate inefficiency potentially. So, you still want to look at the end-to-end, and then you automate and orchestrate it.”
- Beware the governance black hole of citizen development and “shadow AI”: The rise of low-code environments and accessible generative tools means non-developers can build personal productivity tools rapidly. However, without clear visibility from Service Transition, these tools quietly turn into business-critical services with zero underlying support models or strategy. Transition teams must keep active channels with business heads to identify these tools early and wrap proper service design around them before the original creator exits the organisation.
CONCLUSION: REMEMBER THE “WHY”
AI is an incredibly powerful mechanism for expanding practitioner productivity, accelerating baseline design documentation, and freeing humans from administrative drudgery. However, as an active discipline, itSMF members must remember that tool deployment is never the final outcome.
Every successful AI initiative must start with a very simple, uniquely human inquiry: what real user problem are we solving, what is the clear operational goal, and why does this add long-term service value? Everything else is just expensive noise.
The itSMF UK Service Design & Transition Community of Practice meets regularly to share practical insights and support members across the industry. Join our next community roundtable to contribute your voice to the conversation!
Thank you to the following CoP members for their contributions to this discussion:
| Anthony Steer | Lisa Jeffrey |
| Rachael Elliot | Tim Hughes |
| Sarah Routlege | Natasha French |
| Adam Phillips | Bhuvana Sriharimohan |
| Ian Wall | Ray Marshall |
| Helen Nunn | Andy Ferguson |
| Faith Thomas | Chris Miles |
| Jake Saunders | Jacob Ellison |
Chevonne Hobbs
Chevonne Hobbs has worked in IT for 25 years with organisations as diverse as Coca Cola, Leeman Brothers, CAP Gemini, Ricoh and Illuminet Solutions. She is currently Senior Manager Business Consultant at CGI.
