AI bot is resolving 90% of our help desk tickets • The Register

ServiceNow claims it has created an AI agent that is currently solving 90 percent of the inbound IT tickets to the company’s own employee help desk.

ServiceNow said, in addition to using the Autonomous Workforce bot internally, select ServiceNow customers are testing it as well. The firm expects to have the product generally available by the second half of the year.

The internal tickets include high volume issues such as password resets, account unlocks, software access requests, email issues, and VPN connectivity, handled end-to-end within defined permissions and escalation paths.

“In our own environment, over 90% of targeted Level 1 volume is handled autonomously, with resolution rates above 99% for those categories and materially faster than human-only workflows,” a ServiceNow spokesperson told The Register.

It accomplishes this by having the Autonomous Workforce operate on top of the live configuration management database (CMDB), active workflows, policy engines, approval chains, and real transaction history – all updated in real time every time a ticket closes, a workflow executes, or a policy changes, ServiceNow said.

ServiceNow could use a win as Salesforce is targeting its enterprise ITSM customers with its own Agentforce IT Service product. Salesforce CEO Marc Benioff boasted of poaching five ServiceNow customers during the most recent quarter’s earnings call.

Another challenge for ServiceNow is how the Autonomous Workforce will adapt to customer environments. While ServiceNow says its internal documentation is backed with 20 years of experience, documentation inside real world help desks traditionally has been poor to non-existent.

“The documentation problem is real, and frankly most vendors pretend it isn’t. The reason ServiceNow can answer differently is the two decades of structured data that lives inside the platform itself,” according to group VP for AI products Nenshad Bardoliwalla. “This isn’t a system that reads your Word docs and hopes for the best … The pitch is ‘we’re the control plane that aggregates signal from the tools you already have, and we fill the gaps with structured workflow logic built over two decades.’”

Bardoliwalla said, the Autonomous Workforce uses historic ticket information as the knowledge base when answering questions. So far the system has been tested internally by ServiceNow employees feeding it tickets.

ServiceNow broke those tickets down for The Register by type and subtype – which is how help desks organize and manage tickets – revealing that the Autonomous Workforce solved 90 percent of ticket types related to networking (46 percent), hardware (11 percent), and software (43 percent), as well as the following subtypes: enterprise app access, cloud authentication services (33 percent), collaboration tools issues (13 percent), VPN and network connectivity issues (7 percent), laptop and hardware performance issues (8 percent), and software installs and configuration (6 percent).

“How does it know it got the right answer? Because the outcome is measurable inside the same platform,” Bardoliwalla said. “Did the ticket resolve? Did the workflow complete? Did the approval get the right sign-off? ServiceNow closes the loop in a way that a standalone LLM sitting on top of a SharePoint folder simply cannot.”

Bardoliwalla said ServiceNow knows the device, the user, the identity, the access policies, and the historical incident patterns tied to a configuration item, but critically, he said, the Autonomous Workforce knows when it needs to stop and escalate a problem.

“When the gap genuinely exists, the autonomous worker will know what it doesn’t know,” he said. “This is actually a differentiator: a system that says ‘I can resolve 70% of this autonomously and here’s exactly why I’m escalating the other 30%’ is more trustworthy than one that hallucinates an answer.”

Forrester vice president and principal analyst Charles Betz said if ServiceNow has managed to achieve autonomous execution in the Level 1 help desk, that is a “milestone,” since for years AI in the help desk was used for deflection, recommendation, or faster routing.

“End‑to‑end execution at scale is different; it does legitimize AI as operational infrastructure rather than just a productivity aid,” he said. “So yes, this returns real margin — but not simply by shrinking L1 headcount. The value shows up as faster resolution, fewer escalations, better utilization of skilled staff, and the ability to absorb growth without linear increases in labor. The limiting factor shifts from ‘Can the AI do it?’ to whether the organization has the data quality, workflows, and governance discipline to sustain that higher baseline.”

He said it also opens the door to the automation paradox, which he said is similar to what Lisanne Bainbridge wrote about in her 1983 paper, Ironies of Automation. For the help desk, this means what used to be basic support becomes table stakes and “users surface higher‑order needs once the friction is removed.”

“There’s a well‑known dynamic that kicks in once the low‑hanging fruit is harvested. Level 1 doesn’t disappear — the baseline rises,” Betz told The Register. “Routine, procedural issues get automated away, and what remains are harder, more ambiguous, more cross‑system problems. In systems terms, you get complexity creep rather than pure volume reduction.” ®