By now, it’s safe to assume every MSP has seen some version a similar AI demo, where an agent triages tickets, summarizes incidents, drafts a knowledge base article, or suggests the next best action. The output is usually impressive. The problem comes afterward when the pilot ends, the team returns to its normal workflow and, six months later, the AI initiative is still not meaningfully changing the economics of the business.
The failure is often blamed on the model, but the reality is that it’s usually an integration problem.
AI only creates measurable value when it can operate inside the systems where work actually happens. It’s a common theme I’ve seen permeating MSP AI conversations lately. If an agent can read a ticket and recommend a next step, but a technician still has to open a second system, translate that recommendation into an action, and execute it manually, the expected efficiency gains erode quickly. Yes, the agent did the thinking, but a person still had to perform the action.
To be clear, removing the human from the loop entirely is not the goal. There are plenty of events and actions that genuinely warrant human sign-off before execution. No MSP serious about client trust wants to blindly allow AI to make every decision unsupervised — and no client would want them to.
But, the reality is that the majority of IT actions are probably low-risk and can actually be handled autonomously by AI, and the friction created by the need for human intervention in those cases makes a difference, especially for MSPs. While a single internal IT team absorbing that overhead is inconvenient, and MSP absorbing it across dozens or hundreds of clients creates a significant strain on resources, which translates into margin and profitability compression. An AI layer that only adds recommendations for someone to review doesn’t change the economics much.
That’s exactly what Xurrent is trying to address its Xurrent iPaaS, a built-in integration layer that’s included in every Xurrent subscription rather than sold separately as an add-on. Xurrent is an AI-powered service and operations management platform that helps organizations manage incidents, service requests, changes, and related workflows. It’s new iPaaS is designed to let AI agents and teams connect those workflows directly to systems like ServiceNow, Jira, Okta, Google Drive, and other business tools.
Xurrent is making the bet that AI ROI in ITSM will be driven by whether an agent can reach the systems where real work happens, so they can actually carry out work across environments instead of merely advising on it.
That difference — advisory vs. operational — is the crux of the story. Not every IT action should be automated. Provisioning access, touching production systems, or making changes that are difficult to reverse should still require some level of human oversight. On the other hand, routine, low-risk, reversible work should move faster. Without integration, every action costs roughly the same in terms of time, since each recommendation has to be executed manually.
“Integration is the connective tissue that turns AI from advice into action," said Brian Wenngatz, Xurrent CEO. “Without it, an AI agent is an expensive demo with nothing to act on.”
That’s where Xurrent’s play is designed to increase ROI.
According the Xurrent’s announcement, a Harvard Business Review report found that AI tools running alongside existing workflows fail at six times the rate of tools built directly into the work. That makes sense. If AI can’t actually take action and can only make recommendations, it is likely to create bottlenecks.
For MSPs, the integration issue is even more important that individual IT teams, since they don’t standardize on one enterprise application, but inherit their clients’ choices. One might use Jira, another ServiceNow, a third Okta, and a fourth has its own workflow quirks. All of them expect the same fast service, though., and all of them expect fast service. With Xurrent iPaaS, each of those integrations can happen easily, allowing AI to work alongside human techs efficiently.
The pricing model is also worth noting, since Xurrent is including its iPaaS in every subscription, not as a separate add-on. For an MSP, this is useful because metered integration usage or separate licensing costs can complicate pricing models and eat into margins. Folding integration into the base platform removes another layer of friction between AI promise and operational reality.
Think of it in the context of a critical system that’s beginning to fail. The platform can open the incident, alert the right people, and bring AI agents into the investigation before an engineer joins the call. If external systems need to be involved, the iPaaS layer handles that connection. The human still makes the judgment call where appropriate, but they are no longer starting from a blank ticket and stitching together context by hand across multiple systems. The grunt work happens earlier, is completed faster, and provides clear structure for next steps.
Of course, there are other, less critical scenarios, where AI can actually be authorized to perform actions, entirely removed the human element, ensuring techs are actually available to handle the system that’s about to fail.
That can actually change MSP economics, not because it eliminates the human, but because it reduces the amount of low-value effort surrounding the decisions a human still needs to make.
It’s also important to note that it doesn’t all just happen out of the box. Pre-built integrations still need to be configured and validated in real environments. Xurrent also says a natural-language integration builder is still in development.
That notwithstanding, the first wave of enterprise AI proved that models could understand service and IT workflows well enough to be useful. But, they don’t deliver the ROI that’s been promised — or at least expected. So, we’re on to the next wave, where AI agents can actually take action inside the systems of record without requiring a human to manually bridge every gap — but can, when necessary.
Edited by
Erik Linask
