Voice AI is scaling. Can you see how it’s performing?
More of your customers’ first conversations now happen with an voice AI - all without a human in the loop. That’s the promise of voice AI: instant, natural, always-on service at scale.
However, scale cuts both ways. Every automated call is a decision your business is making on a customer’s behalf, and if you can’t see how those calls are going, you’re flying in blind. Did the voice agent resolve the issue? Did it reply fast enough to feel natural? How did the customer feel during the call?
Voice is its own discipline compared to chat-driven experiences. A delay of a couple seconds that’s tolerable in chat becomes an awkward silence on a voice call. Additionally, voice conversations unfold dynamically across different use cases – an identity can be meticulously verified out loud character-by-character, frequent interruptions must be managed, and tool actions may be fired mid-speech. Measuring voice well means measuring all of it: the outcome, the speed, the sentiment, and the tools working underneath.
ServiceNow voice analytics is built to provide clarity and enable key business insights on what happened for every voice interaction. These analytics live inside Assistant Designer, the same place you build and test your voice AI experiences. Four views (Overview, Performance, Insights, and Assist consumption) are available to provide a comprehensive picture of voice AI performance.
Overview: Know whether your calls are getting resolved
Note: All screenshots in this article show sample data from a test environment. The figures are illustrative and not indicative of expected performance.
Start with the question that matters most: is your voice AI actually helping? The Overview tab answers it at a glance. It tracks the essentials – including total conversation volume, resolution, handle time. You can also view resolution by individual agent, so you can see which use cases are winning and which are stalling.
For example - say your AI voice agents normally resolves 80% of their calls. This week it drops to 50%, and more callers are being handed off to live agents. With voice analytics, you can view which specific agents are driving the performance dip and also review interactions for calls that failed. They break at a similar point: the agent won't file a ticket until the caller reads out an ID number they rarely have on hand, so they give up and ask for a person. Now the fix is clear too: modify your voice agents to stop requiring that ID upfront, or let the agent look callers up another way. That's the insight which resolution provides.
Resolution is inferred generally by AI from the conversation itself, so you learn whether the agent is working without waiting on survey responses that most customers never send.
Performance: Find what delays your calls (or breaks them)
Performance shows you the mechanics: how fast the agent responds after someone stops speaking, how long its tool actions take and how often they succeed, and how smoothly it verifies a caller’s identity (authentication).
Averages hide the worst moments, so these numbers are reported as percentiles. An agent can average a ~1 second response and still leave one in ten callers waiting through a five-second silence. That slowest slice is the call that feels broken, and percentiles enable visibility into these edge cases.
Insights: Learn how customers really felt
While numbers may inform you what happened, they don’t tell you how calls felt. Insights analytics reads the conversations themselves, using LLM analysis of call transcripts to surface key sentiment signals – such as whether the caller expressed frustration or confusion during the interaction. Because it runs on every transcript instead of an opt-in survey, you hear from the silent majority who would never fill out a form but make their feelings perfectly clear on the call.
Assist Consumption: Know what it costs, not just what it does
Performance is only half the story. The other half is what all that automation costs to run. This Assist Consumption view gives you that visibility: your overall spend (assists consumed), how it’s trending, and which agents are driving the most cost - so you can plan capacity and spend with real numbers instead of guesswork.
Put that next to your resolution, performance, and sentiment numbers, and you can finally give every business leader the answer they're really after: proof the AI is worth what it costs.
Turn every conversation into your next improvement
Every call your voice AI handles is telling you something: what customers need, where the experience breaks, what to build next. Analytics is how you hear it, at the scale voice AI operates.
Try it today: https://www.servicenow.com/docs/r/conversational-interfaces/now-assist-in-virtual-agent/voice-assist...
