Quality assurance for your AI voice agents
Human QA scorecards do not fit voice bots. Empathy and grammar are the wrong questions; whether the bot let a caller interrupt, how long it paused before responding, and whether it invented information are the right ones. Xperia scores bot-handled calls on their own criteria.
Conversational fluidity
Barge-in handling — does the bot stop talking when the caller interrupts. Turn-taking, so the two parties are not speaking over each other. Filler sensitivity, dynamic pacing, response latency, and how smoothly it transitions between topics or hands off.
Emotional design
Tone matching against the caller's state, and conciseness — whether the bot answers the question or delivers a paragraph when a sentence would do.
Acoustic quality
Volume normalisation, speech-to-silence ratio, voice clarity, phonetic accuracy on names and identifiers, and prosody and intonation.
NLU intelligence
Intent recognition accuracy, entity extraction, fallback behaviour when the bot does not understand, context retention across turns, and hallucination — whether it stated something it had no basis for. For most teams this last one is the reason to measure at all.
Bot and human calls, side by side
Bot-handled calls are segmented from human calls throughout the product — analytics, call history, upload and parameter configuration — so each is scored on its own criteria while both stay visible in one account.
Questions
Frequently asked
Can Xperia evaluate AI voice agents, not just human agents?
Yes. Xperia scores AI voice-bot calls against 18 dimensions designed for automated agents, grouped into conversational fluidity, emotional design, acoustic quality and NLU intelligence. Bot calls are segmented from human calls across the whole product.
What does Xperia measure on a voice bot?
Barge-in handling, turn-taking, filler sensitivity, dynamic pacing, latency, tone matching, transitions, conciseness, volume normalisation, speech-to-silence ratio, voice clarity, phonetic accuracy, prosody and intonation, entity extraction, fallback handling, intent accuracy, hallucination and context retention.
Can Xperia detect when a voice bot hallucinates?
Yes. Hallucination is one of the scored NLU dimensions, alongside intent accuracy and context retention, so you can track how often a bot asserts something it had no grounding for.
Can we score human agents and voice bots in the same account?
Yes. Bot calls are separated from human calls throughout analytics, call history and configuration, each scored against its own criteria, without needing a second account.
Related
AI Call Scoring
Xperia scores every call against 48 quality dimensions across 10 categories. Weightings, pass criteria and the parameters themselves are configurable per department.
AI Assistant
Xperia's AI Assistant answers natural-language questions about your calls, agents and quality scores, returning tables and summaries instead of dashboards to navigate.
Try it on your own calls
Every account starts with $3 of free credit — around 75 minutes of call analysis. No card required.
Start Free