Voice AI for the IT Service Desk: How It Works in ServiceNow

Traditional service desks face a persistent bottleneck where end users wait on hold for simple inquiries, while support agents deal with endless queues of repetitive calls. Voice AI for the Service Desk is built to address this issue.

In our recent webinar, Astreya's Service Desk Practice Lead Josh Spring joined Guillaume Seynhaeve (ServiceNow Practice Lead, 3CLogic) and Ajay Goel (Sr. Customer Success Manager, 3CLogic) to show how it works inside ServiceNow. Lauren Frazier, Astreya's VP of Marketing, moderated. The session included a live call with an AI agent, a look at the admin configuration, and an unscripted Q&A.

Summary: Voice AI for the IT service desk uses conversational AI agents to answer calls, verify the caller, capture intent, and handle routine requests like incident creation and status checks. It escalates to a live, human agent, with full context. When it's built natively on ServiceNow, calls, tickets, knowledge, and agent workflows all stay in one system of record.

‍

Why is voice still the most important service desk channel?

ServiceNow has pushed IT organizations toward portals, chat, and self-service. However, voice is the channel people use for urgent issues or when there is an impasse in the portal.

Josh described the gap Astreya saw in its own practice: “ServiceNow serves the portal channel and the chat space well, but voice is somewhat of an open area."

Guillaume explained why that gap matters more than volume alone suggests:

“An IT service desk, for all intents and purposes, is what keeps the heartbeat going. Everything's fine until it's a fire drill … When people call, those are your most high-value conversations because people are calling because the problem is severe enough, by their definition, to justify one."

What can a voice AI agent do on a ServiceNow service desk?

The panel focused on predictable, repetitive, and transactional requests. Josh listed the most common use cases:

• Fast escalation by reaching a live agent the moment a person is needed.
• Ticket creation and request submission by voice.
• Status checks on an existing incident or request.
• Known-issue lookup so a caller can tell whether their problem is part of a wider outage.
• Knowledge article delivery, where the agent walks the caller through a fix or sends them the self-service article.
• After-hours intake, in which Guillaume described having the case created so that, when a team logs in the next morning, all the information is already gathered and the workflow is already kicked off

How does voice AI work inside ServiceNow?

Guillaume walked through both the admin build and a live call:

Administration lives inside ServiceNow

3CLogic installs through an update set that appears in the ServiceNow menu. Organizations manage live agents and AI agents in the same place. That matters to Josh, who said he wanted a platform he could “(hand to) my operations leader or my supervisor … without having to be a full-fledged rocket scientist."

Skills-based routing is configured as well. In a managed services environment, agents have different proficiencies, and Guillaume said that the goal is getting callers “not just to the next person, but to the next person or agent that can help you."

Two layers: Deterministic flows plus AI agents

The architecture has two tiers:

1. An interaction manager. This is a deterministic workflow, similar to an IVR, that controls the opening experience and handles steps that always run the same way.
2. Voice AI agents attached to that flow. These use large language models to hold more natural, conversational interactions.

In the voice agent builder, each agent gets a job description, dos and don'ts, a personality, supported languages, and a set of abilities. These are the actions it can take in ServiceNow. The demo Voice AI agent, Isabel, gathers caller details to personalize the call, looks up the latest comment on an incident, and opens new tickets.

AI agents can also be limited to a specific knowledge base, so they only answer from the articles your team approves.

‍

The live call: Verify, understand, create, and escalate

During the demo call, Guillaume told Isabel he was locked out of his computer and had already lost portal access. Isabel greeted him and asked for his six-digit employee ID. She then clarified the issue, offered the password reset path, and transferred him to a live agent once it was determined that one was needed to resolve the issue.

Before the transfer to a live agent, Isabel opened a new incident with a short description and identified the caller using his phone number and employee ID verification.

The handoff: Live agents get a headstart

On the live agent side, the call begins with a lot of the groundwork already done. They can see that the end user has been verified and have access to a summary of the issue, along with a full transcript of the interaction with the AI agent. . If the AI conversation is long, the agent can use ServiceNow's AI to summarize it.

Automated post-call work

When the call ends, the transcript and summary automatically post back to the interaction record.

‍

Should voice AI replace live agents on the service desk?

The panel agreed it shouldn't. Ajay framed it this way:

“Instead of trying to compare or pit AI agents versus human agents, organizations get the best result when they unify the setup that connects both AI and live agents seamlessly … AI takes care of the upfront work — things like verifying the caller identity or capturing the intent — and then it can pass that intent to the human agents who are ready to solve complex problems."

Guillaume drew the line between the two:

“We see voice AI as very useful for transactional, repetitive things … You want to reserve your live individuals for things that are more nuanced, that require trust, accountability, judgment. And so you need a platform that can juggle both."

In practice, 3CLogic routes AI agents and live agents through the same place. Guillaume summed it up: "You have to be able to hold your voice AI agents to the same standards as you hold people."

How long does it take to implement voice AI for a service desk?

Josh estimated 10 to 12 weeks, depending on the complexity of the environment. He said it starts with the experience, not the technology:

“It really starts with the conversation about what sort of experience do we want to create. What do we want this to look and feel like to an end user as they engage?"

Below is an example configuration:

1. Experience design. Defining how the call should look and feel to the caller by using what worked or did not work in the organization’s past phone system.
2. Call flow. Josh called this the "plumbing" that makes the experience reliable, informed by what worked (or didn't) in the previous phone system.
3. Pick the top 5 to 10 use cases.
4. Measure and iterate. Josh's preferred metric is customer effort score. “Are we actually producing an easier-to-use experience? Is this easier than it was before?"

‍

What slows voice AI implementations down?

The panel pointed to the following three areas:

1. Data quality. Ajay noted that the foundation depends on how organized and granular your data is. Guillaume agreed that the experience will be limited by the quality of your ServiceNow installation.
2. Every service desk is different. “Everyone's a variation of a snowflake," said Guillaume. The goal isn't to change your processes, “but help bring those up to that voice-level engagement," so callers feel like they're dealing with the same organization.
3. Unclear standards and expectations. “You can't expect voice AI to do anything consistently that your team isn't doing consistently," said Josh.

‍

‍

How do you measure voice AI agent performance?

Classic voice metrics like abandonment rate and speed of answer still apply, but the panel said they aren't enough. The metrics discussed include:

• Containment rate: How much of the workload the AI actually handles
• Resolution and accuracy
• Sentiment: Positive, negative, or neutral
• Adherence to guidelines: For example, whether the call opens and closes the way your organization expects
• Customer effort score and CSAT
• Cost per ticket and speed of service

3CLogic's tool lets your organization define its own criteria, much like QA scorecards for human agents. You can score AI agents at scale and track them across versions. Guillaume said this visibility is what gets projects into production:

"The reason why most projects don't leave the pilot phase is because there's a concern that there's no audit trail, no visibility. It's a black box once you let it loose."

Josh added the operational angle, that anyone who has manually audited service desk calls knows it's a lot of work and a lot of operational overhead, so automating QA is a significant saving on its own.

[SCREENSHOT ~57:20: AI evaluation scoring dashboard]

What should IT leaders look for in a voice AI platform?

Josh's reasons for choosing 3CLogic double as a checklist for any evaluation:

• Capabilities that aren't "overbuilt or overpromised." It should do what the vendor says it does.
• It reuses what you already have by including existing ServiceNow workflows, knowledge articles, and SOPs.
• AI built into the core platform, not added on.
• Low admin overhead.
• A partner that aims for your independence.

What's next for voice AI on the service desk?

Guillaume expects service to move from separate channels to single, multimodal AI-powered experiences, with voice as the main channel. Mixing voice with digital inputs makes calls faster and expands what a voice AI agent can resolve.

Watch the full session

See the live demo, the agent handoff, and the full Q&A in the on-demand recording.

Astreya's service desk AI agents are available on the ServiceNow Store. To see how voice AI fits into a managed service desk, explore Astreya's Service Desk services or our AI & Automation practice.
‍