AI for IT Operations: Why Most Tools Fall Short (And What Changes When They Don't)

AI for IT operations, also called AIOps, refers to the use of artificial intelligence and machine learning to automate, monitor, and improve how IT teams manage their environment. In practice, it covers everything from surfacing device compliance issues to triaging access requests to identifying unused software licenses. The goal is to reduce the manual work IT teams do correlating data across tools and give them more time for the decisions that actually require human judgment.

That's the promise. The gap between the promise and what most teams are actually experiencing is significant, and it has a specific cause.

What Is AI for IT Operations?

AI for IT operations describes technology that applies machine learning, natural language processing, and automation to the day-to-day work of IT administration. The term is often used interchangeably with AIOps, though it's worth noting a distinction: AIOps traditionally refers to AI applied to infrastructure monitoring and event correlation, the kind of work DevOps and platform engineering teams do. AI for IT operations, in the workplace technology sense, covers a broader scope: managing employee devices, access provisioning, identity management, SaaS license governance, and IT support.

Both applications share the same underlying premise. IT environments generate enormous volumes of data across many systems, and humans spend too much time manually connecting the dots between them. AI is supposed to do that correlation work faster and more reliably.

The technology has improved substantially. The problem most IT teams are running into is not the AI itself.

Why AI for IT Support Hasn't Lived Up to the Hype

Most IT admins have encountered AI for IT support in some form already. A chatbot that deflects help desk tickets. A reporting tool that generates outputs a few minutes faster. A log summarizer that still requires a human to interpret the output.

As Zaafir Kherani, Product Lead for Rippling IT, put it: "The core issue is fragmentation. Most IT environments were not designed as one system. They were assembled over time."

That assembly problem is what makes AI for IT support so difficult in practice. An identity provider here, an MDM solution there, a separate access management tool, a ticketing system added later. Each one works fine on its own. None of them share context with the others.

So when a simple question comes in — "Why can't this employee access this app?" or "Which devices in my fleet aren't encrypted?" — the answer isn't in any single tool. An IT admin has to pull from multiple systems, cross-reference manually, and piece together a partial picture. If that answer requires a fix, the whole process starts over in a different tool.

Carter Francis, a 20 year IT veteran who is a former IT Manager, described what that looks like in practice: "Your answer isn't just in one place. Your answer is all over the place. And you, your brain, your process is all over the place trying to figure it out. So you're translating everything and then you have to also translate it for the user."

That translation overhead is what AI for IT support is supposed to eliminate. Most of the tools on the market today haven't solved it because they're sitting on top of the same fragmented environment that created the problem in the first place.

Three Reasons AI IT Operations Tools Fall Short

When AI shows up in an IT environment today, it typically fails in one of three predictable ways.

Lack of environmental context. Most AI for IT operations tools don't actually know your environment. They don't know who your employees are, which devices are assigned to them, what access they've been granted, or which policies govern that access. Without that foundation, any answer the AI produces is either a generic pattern match or a guess. Neither is acceptable when you're the one accountable for what happens next. Fast wrong answers are worse than slow correct ones.

Answers that can't be verified. Even when an AI produces a response that looks right, tracing it back to a real policy in a real system is on you. If the AI says a user has admin access because of a particular policy but you can't confirm that in the actual system of record, you still have to go check manually. That's not a productivity gain. It's an extra step.

No path to action. This is where the gap is most costly. Even if an AI correctly identifies that a device has fallen out of compliance or that a former employee still has access they shouldn't, most AI IT operations tools today cannot safely do anything about it. They surface the problem and stop. The fix still requires the same manual workflow IT was running before: open a ticket, find the right person, wait for a response, confirm the action, close the loop.

Zaafir Kherani put the dependency plainly: "If you don't have context and trust, the last thing you're going to do is volunteer to plug something random into your MDM."

What AI for IT Operations Actually Needs to Work

The condition that changes all three failure modes is the same: a unified system of record where identity, devices, access and permissions, policies, and employee data all live together.

When that foundation exists, AI for IT operations can reason over actual relationships rather than fragments. It already knows which team an employee is on, which devices are assigned to them, which apps they can access, and which policies govern each of those things. When you ask it a question, it traces through real data rather than synthesizing guesses across disconnected sources.

That changes what "verifiable" means. You're not evaluating an AI inference. You're looking at the actual state of your environment in the system where your policies are defined. If the AI says a request was blocked by a specific policy, you can click through and see that policy. The answer is auditable by design.

Once the AI's answers are grounded in a unified system of record, they're verifiable, allowing you to trace any output back to an actual policy or device state. That same foundation is also what makes action safe: the AI operates within your existing governance structure, stages proposed changes for human review, and can't act outside the permissions you've already defined.

CJ Hughes, Global Head of IT and Security at Omniscient, described the expectation IT admins have for how this should work: "Best case would be to check out what it was going to do, and demonstrate it before it applied it."

That preview-before-action model is the right frame. Not AI operating autonomously, and not AI that merely describes problems. AI that does the diagnostic work, proposes the solution, and waits for a human to approve.

AI for IT Operations Use Cases: Four Real-World Examples

The practical difference between AI for IT operations that has full system context and AI that doesn't is easiest to see in specific workflows.

New hire readiness

Getting a complete picture of who's starting next week normally means checking the MDM for device shipping status, confirming with HR who's on the onboarding list, and logging into each application to verify access provisioning. That's three tools, three logins, and a manual reconciliation. With full system context, you ask one question: "Who's starting next week, what apps are they set up for, and has their gear shipped?" One answer, pulling from all three sources. The automated onboarding workflow runs correctly because the AI is operating from the same data that drives provisioning. At o8t, a neurotech company managing employees across Australia, the US, and Canada, each new hire previously required three hours of manual laptop configuration. After consolidating onto a unified platform, that configuration ran automatically. Stephen Lai, VP of Finance, put it plainly: "With Rippling, everything is there. Across different modules, you can log in and see what you need to see." The result: 10 hours saved per month across IT and payroll, with zero additional IT headcount added.

License reclamation

Unused software seats are a quiet, consistent drain on IT budgets. Finding them normally means exporting usage reports from each application's admin console, matching that against active headcount, and manually building a picture of who's actually using what. With context-aware AI for IT operations, you ask "find anyone who hasn't opened Zoom in the last 60 days," get the list immediately, and reclaim, downgrade, or reallocate seats on the spot. What used to be a half-day exercise becomes a five-minute decision.

Lost device response

An employee reports a missing laptop. In a fragmented environment, that's a ticket, a device lookup in inventory, an ownership confirmation, and a remote lock command through your MDM, assuming you have direct MDM access and know where the device is registered. With full system context, you type "lock Erica's laptop," the system asks which Erica you mean, you confirm, and the action is staged for execution. One click. The IT offboarding process follows the same principle for departing employees, making it possible to close access and secure devices in a single workflow rather than across multiple platforms.

Fleet health at a glance

Diagnosing the health of a distributed device fleet normally means pulling separate reports from your MDM, sorting by device age, checking battery reports, and cross-referencing serial numbers against employee records — a half-day exercise that most admins run reactively, after something breaks. With AI that has full system context, you type one question and get an immediate fleet-wide view: battery life, storage usage, OS versions, and which devices need attention. Bob Parsons, IT and Cyber Security Specialist at Ghost Robotics, describes what that shift looked like: "You go from problem to problem and ticket to ticket. But being able to pull this data and look further down the road and plan better. It was a huge help." For Ghost Robotics, the result was a 35–45% reduction in repeat tickets and an average ticket resolution time of 10 minutes. "It gives you so much more time to focus on infrastructure and critical projects and future planning that you just don't have without this tool," Parsons said. "Let's not react, let's be proactive." Carter Francis noted what this shift means for end users as well: "It's not help desk deflection. It's giving users reliable answers and actual action without needing to step in every time."

What to Look for in an AI IT Operations Platform

Not all AI for IT operations tools are built the same way, and the SERP is full of vendors who use "AIOps" to describe very different things. When evaluating options, a few questions cut through the noise quickly.

Does the AI have access to the full environment, or is it reading from a subset of your data? An AI tool that connects only to your MDM cannot answer questions about access policies. One that connects only to your identity provider cannot tell you about device compliance. NIST SP 800-124, the federal guidance on mobile device management, outlines the full scope of what device management should cover, and most AI tools only see a fraction of it. The intelligence of the AI matters less than the completeness of the data it has to work with.

Can you verify the AI's answers without leaving the platform? If confirming an AI-generated answer requires you to log into a separate system, the tool hasn't solved the fragmentation problem. It's added a layer on top of it.

Does the platform support human-reviewed action, or does it operate autonomously? IT teams are not looking for AI that acts without approval. The value is in eliminating the investigation time, not the human decision. A platform that stages proposed actions for review is more useful, and more trustworthy, than one that either only reports problems or takes action without a confirmation step.

Is the AI's output auditable? For compliance purposes and for practical accountability, IT admins need to be able to trace what the AI did, why, and when. Platforms that support this natively are the ones that hold up when security teams or auditors start asking questions. This is especially relevant as shadow IT and shadow AI from employee-adopted AI tools becomes a growing exposure: admins need the same audit visibility over AI-initiated actions as they do over human ones.