Is your service desk actually ready for AI?

Right now, everyone’s feeling the pressure to increase the amount of AI they’re using in their working lives. The potential of AI is significant, and everyone from the service desk up to the boardroom wants more AI because of how it can improve productivity. The C-suite wants the AI strategy, managers want visible, measurable efficiency, and suppliers are pushing AI transformation, so working out how to get the most from your AI deployment is paramount.

In this guide, we’ll help you figure out if your service desk is actually ready for AI. Read on for practical tips on how to work out your AI readiness, and for concrete next steps you can get started with today.

Why AI fails

The pressure to implement AI is real, and that’s because its potential is real. Unfortunately, this desire can have counterintuitive results, leading to teams deploying tooling before the foundations are properly in place. As we’ll see, you can’t expect to have solid results on shaky foundations.

At the end of 2025, Gartner surveyed 782 infrastructure and operations leaders and found that "only 28% of AI use cases in infrastructure and operations (I&O) fully succeed and meet ROI expectations, while 20% fail outright" — and 38% said "poor data quality or limited data availability was a direct cause of AI project failure."

Issues such as these don’t necessarily suggest that AI has no inherent value. Instead, it suggests that the quality of the fundamentals impacts the level of return on investment: strong foundations likely lead to positive ROI.

Introducing the Foundation-First Framework

So, how can you know if your company is ready to take full advantage of AI tooling? By running a full examination of your current state of play.

Introducing the Foundation-First Framework. In the next few sections of this article, we’ll talk you through the Foundation-First Framework, and show you how you can use it to understand whether AI will add any real value to your team.

The Foundation-First Framework works by asking you to concentrate on a single use case, and then drill down into three key foundational pillars:

• Your knowledge
• Your process
• Your data

Any AI that you look to introduce will need to be built on top of these three areas. The quality of these three pillars determines the quality of the output. Think of AI like an amplifier: good stuff in means good stuff out; bad stuff in… well, you get the idea.

Quick side note before we dig into things: it’s important to state that the Foundation-First Framework is not another maturity model. Instead, it’s a practical lens helping you to validate your potential use cases, to see whether you’re ready to introduce AI to your service desk.

Knowledge: owned, findable, and fresh enough to trust

Let’s start with knowledge. There are two main criteria for how functional your company’s knowledge base is currently. The first is ownership, for example, who actually owns your knowledge? Let’s be honest, “everyone owns the knowledge base” is really just a polite way of saying that nobody does. Without accountability, how do you know what the single source of truth is? Which of three knowledge items on password resetting is the right one?

The second point is findability. If the right answer can’t be found by a person, then it won’t be found by AI, either, because the same rules apply for humans and machines. If the answer is in your team’s veteran’s head, or if it’s in some obscure private folder called ‘final_v2_REALLY_final’, then maybe sort that out before scaling your knowledge base using AI.

Test 1: If you want to work out how findable your knowledge is, ask a new colleague to find it. If they can’t discover it, then AI will struggle, too.

A map of who owns your knowledge—and where to find it—is not enough in itself. The knowledge also needs to be current enough that it’s still accurate (because feeding AI outdated knowledge can be actively harmful); it needs to cover all of the common questions and repetitive responses that make up the bulk of your requests; and it needs to be a feedback loop—in the sense that the new knowledge generated for every solution for every request is fed back into your knowledge base.

If you can get these essentials right, then when it comes to adding AI to your workflow, you’ll be generating accurate responses that are genuinely helpful to your users.

Process: can the work actually be followed?

The second foundation is your process. Whether or not your process is ready to scale up using AI is basically down to whether you can answer one simple question: can the process actually be followed?

Follow these four checks to learn the answer:

1. Steps. Is your process documented well enough that somebody new could follow it?
2. Handoffs. When a process moves between different people or teams, does the ownership stay clear? Is all information moved between groups without any losses?
3. Exceptions. Are the edge-cases known and documented?
4. Review. Do you understand what parts of the work you’re happy for AI to run with, versus where you want a human to make a decision?

Test 2: Ask two experienced team members to handle the same request. Do they come back with the exact same approach? If they do, great, that process is ready to automate. If they don’t, then it’s time to clarify the approach.

Data: can you trust it?

The third and final foundation is your data. Putting AI on top of bad data means whatever process is scaled up or automated can’t be trusted. But what is data readiness? It’s certainly not about perfect data. As with most things in life, perfect doesn’t exist. AI basically needs data that’s consistent, permitted and useful enough to learn from.

‘Useful enough to learn from’ probably means that your categories are spelled correctly, that key fields are complete enough and that outcomes are described in enough detail to be helpful. ‘Safe enough to use’ probably means that access rules are clear, your sensitive data is handled with intentionality, and where necessary, human oversight is well-defined.

One quick tip: the EU AI Act also treats data quality, documentation, traceability and human oversight as essential. Get these right once, and you'll have data your AI can learn from and a head start on compliance, all from the same piece of work.

The same foundation, two outcomes

Knowledge, process and data are the three foundational pillars that will either make or break your AI deployment. If you’ve got weak foundations, AI will take your tasks and confidently give you the wrong output.

Think of a repetitive task you'd like to hand off to AI. Now imagine running it on weak foundations and see what happens:

• Outdated articles: the AI confidently shares a fix for software you retired last year.
• Inconsistent categories: tickets land with the wrong team because "network" means something different to everyone.
• Unclear escalation: urgent issues sit in the queue because the AI doesn't know who should pick them up.

Your team still gets answers, but they're inaccurate often enough that nobody trusts the results.

On the other side, consider what happens when running the same scenario on strong foundations:

• Summarize: forty updates become one clean summary so that nobody has to ask the user to explain it all again.
• Classify: categories and routing become consistent so requests reach the right team the first time.
• Draft knowledge: solved requests become draft articles.
• Communicate: replies get clearer, and automated translation stops slowing things down.
• Self-service: people get a useful answer before they even submit a request.

Each of these wins automates a specific, repeated and annoying task, freeing up your teams to work on the hard problems they might not otherwise have the time for.

And these are only the first wins…

Assistive versus autonomous AI

The sorts of tasks we’ve been considering so far have been assistive. But what about agents? Agentic AI is autonomous. But assistive AI can operate with a few small cracks in the foundations, because a human is always going to be in the loop. But that’s not the case when it comes to agentic AI.

This means that if you’ve got absolutely perfect foundations, you can use agentic AI. But if you don’t, then this limits the AI you can deploy to assistive-only. The more work AI does on its own, the stronger knowledge, process, and data need to be.

Start with one repetitive task

The pressure to build AI into your workflows is probably going to keep increasing. That’s because the potential of AI is massive—but only if the right foundations are in place to ensure that what comes out the other end is actually helpful and trustworthy.

So, start with one repetitive task, follow the Foundation-First Framework, and let us know how far you get. Or, if you’d like, talk to us about a readiness check: Get Ready for AI.