Like every other business, IT service providers are racing to adopt AI. But, compared to other types of organizations, service providers face some unique challenges here. Obviously, the technical side of AI adoption is hugely important, but it’s only part of the story. Instead, for service providers, many of the biggest blockers come from what we could call “soft” factors.
This includes things like expectation management, internal buy-in, or the level of understanding of AI and what it can do within client organizations.
Maybe the most important of all is trust. In particular, around how safe, effective, and predictable AI is.
While there are technical levers we can pull to establish trust, a huge part of the puzzle for service providers is allowing their clients to see AI working and delivering value in familiar, low-risk, and often fairly boring tasks.
AI’s trust problem
So what do we mean by trust when we’re talking about AI?
There are a few components to this. At a purely functional level, there’s confidence that AI-powered systems will make the right decisions within workflows. Not just that, but that they’ll consistently make the right decisions. This isn’t something that we generally need to deal with when adopting deterministic automations.
It’s also important to tie this to risk tolerance. That is, in terms of the consequences if an AI system gets something wrong. This adds a situational element to trust. A particular client might trust AI with certain data and workflows, but not with others - potentially even if solutions exist that meet their requirements on paper.
We can also think of trust outside of individual workflows and in terms of AI as a whole - especially compared to deterministic automations. With a traditional automation, such as a script or a workflow builder, we can essentially always know what the system is doing, and how, and why. By extension, we know what is happening to our data.
This isn’t quite as straightforward with AI-powered systems.
The issue here is transparency. We can tell a model what we want it to do, but we don’t always know exactly how it does this or where the data and context go. While this is solvable for sensitive use cases by using locally hosted models, observability tools, or other measures, it still contributes to AI’s overall trust problem.
Why is this such a blocker for service providers?
Trust is often a bigger blocker for service providers than for other types of organizations. In large part, this comes down to the fact that we’re dealing with external clients who can have highly varying levels of internal technical expertise. At the same time, IT services are mission-critical for most organizations. Service disruptions can easily mean the whole company downing tools.
So, it makes sense that at least some clients would be wary of using solutions they may not fully understand for some of their most important internal processes.
At a more mundane level, service providers also have to consider how their clients perceive AI adoption. Specifically, many clients may be worried about whether AI-assisted delivery, recommendations, configurations, or workflow changes are reliable, explainable, and properly governed. While a range of governance and auditing tools exist, the burden is still on service providers to prove that agents are reliable and effective.
Big promises don’t help
Unfortunately, the broader discourse around AI isn’t always helpful here. This can skew expectations in a few different directions. Remember, most service providers’ clients aren’t tech people. So, when people hear big sweeping claims about AI, a certain proportion will take this to heart, even if the reality of what AI adoption means for their internal workflows is quite a bit more mundane. In other words, clients might not have fully realistic expectations about what’s possible or even desirable here.
On the flip side, this can also fuel skepticism and a lack of trust. If your clients are constantly being bombarded with sales pitches about AI that don’t seem plausible, it hurts trust, even for discrete, highly achievable use cases.
While these might seem like two separate challenges, the solution to both really comes down to education and proof.
Or to put it more concretely, we need to help our clients understand what AI actually achieves within their workflows and show them the real-world results.
Trust is built in small steps
Trust is earned, even for technology. In practice, this means that IT service providers are often better off focusing on relatively small, easy-win use cases in order to secure buy-in from their clients.
But it would be wrong to think that this means compromising. The thing is, even ignoring all of the dynamics between you and your clients, most service providers should be prioritizing fairly mundane use cases anyway, at least in the initial stages of their AI adoption journey.
Often, the biggest and fastest impact that we’ll see from AI adoption is for basic tasks within workflows. Things like routing, approvals, triage, categorization, escalations, self-service, and handling SLAs, which can create a lot of friction in workflows but aren’t actually that complicated.
These kinds of tasks can be difficult to handle with deterministic automations. So, until relatively recently, they’ve taken up huge resources to deliver. They’re also some of the best examples of real, discrete business problems that can be solved with AI agents. Specifically, natural language processing and agents that are able to autonomously invoke tools and functions are enabling teams to automate time-consuming tasks that would otherwise require manual human intervention.
So, we can prove value relatively easily. In turn, this builds trust.
Just as importantly, most of the use cases we mentioned a second ago actually give agents quite limited autonomy. For the most part, the decisions are limited to things like which human to route a submission to or which one of a small set of defined actions to take.
As such, these kinds of simple, almost boring tasks are crucial for IT partners that want to build trust in AI systems because they’re highly boundaried and keep human users tightly in the loop, while still delivering value for clients.
How IT service providers can turn this into an adoption roadmap
So how do we put this into practice?
One key mistake many teams make is starting with the idea that they want to utilize AI more and then going looking for use cases. Often, it’s more helpful to think about this the other way around by identifying low-risk, high-friction tasks. For example, where the decision logic isn’t complicated, but pulling out the necessary information is. Then we need to figure out the boundaries. That is, what the AI can do, what data it needs, what it can’t do, and where a human steps in.
Early on in our adoption process, it’s also useful to trial agents that solely provide recommendations to human users for final approval. This can then be expanded into fuller autonomy later.
Take ticket routing. We don’t need to start by letting an agent resolve issues end-to-end. A better start is using AI to read an incoming ticket, identify the affected service, suggest the right queue, and explain the recommendation. A technician still approves the route, but the client can now experience the AI working within clear limits.
Finally, throughout this whole process, it’s crucial to measure, iterate, and refine in order to build confidence in our solutions. That is, the more accurately we can track resolution times, time saved, and client feedback, the more easily we’ll be able to prove value and build trust for scaling our AI adoption.