​

Conversation and Call Intelligence Landscape

Four analysis surfaces in the Orbit console all work the “intelligence” territory: Voice → Intelligence, Insights → Topic Intelligence, Insights → Sentiment, and Voice → Voice Agent Quality (VAQI). They share vocabulary (sentiment, topics, intents) but slice different traffic, at a different unit, on a different cadence. This page names the question each one answers, puts them side by side, and walks the three questions operators ask most.

​

The four surfaces

​

Voice → Intelligence: what is happening inside our calls?

Per-call post-call analysis of your recorded, transcribed voice calls, at Voice → Intelligence (/voice/intelligence). It rolls the analysis for the selected window into a sentiment timeline, a keyword cloud, topic clusters, a call-intent breakdown, and conversational-dynamics metrics, and every spike drills down to the exact calls behind it. Voice-only, read per call. The speech-analytics categories and intelligence operators you define here are tenant-owned controls over what the post-call pass computes.

​

Insights → Topic Intelligence: why do customers contact us, and is it rising?

Cross-channel contact-reason rollups at Insights → Topic Intelligence. Every conversation across voice, SMS, WhatsApp, email, and web chat is bucketed into its contact reason and ranked by volume, with a trend direction against the prior equal window, a per-channel split, and per-topic CSAT, deflection, and handle-time outcomes. Read it for the portfolio view: what drives inbound load across all channels.

​

Insights → Sentiment: is the customer experience trending up or down, and where?

The cross-channel sentiment portfolio at Insights → Sentiment. Per-message sentiment scores roll up into a net sentiment index with by-channel, by-agent, and by-resolution breakdowns. Read it for the executive read: one number for the whole experience, with the panels localizing a dip to a surface, a roster, or a resolution state. Sentiment scoring itself is a tenant-owned control under Settings → Compliance → Inbox AI privacy.

​

VAQI: is the voice agent holding up its side of the conversation?

Per-session mechanical quality of voice-agent sessions, at Voice → Voice Agent Quality (/voice/vaqi): latency percentiles (TTFB, turn gap, and the STT/LLM/TTS component split), turns per session, barge-in rate, and barge-in success. VAQI never reads content; it measures how the line behaves. Read it when the complaint is about speed and flow, not about what was said.

​

Side by side

VAQI offers 24h, 7d, and 30d windows; the other three add 90d. Use 24h for incident triage and 30d or 90d for decisions where one bad afternoon should not move the number.

​

Pick this when

​

Diagnosing a call-quality regression

Start at VAQI. A step change in p95 TTFB or turn gap, isolated to one component card (LLM, STT, or TTS), is a mechanical regression and usually anchors to a deploy date. If VAQI is healthy but callers are still unhappy, move to Voice → Intelligence: a sentiment dip with a clean start day points at content (a script, an IVR prompt, a routing change) rather than mechanics. A latency regression belongs to VAQI; a sentiment shift belongs to Intelligence.

​

Spotting an emerging contact driver

Open Topic Intelligence and read the trend column, not the volume column: a small topic marked rising is the early warning that a large flat topic hides. Confirm it in the per-channel split, then take the voice side of it to the Voice → Intelligence topic clusters when you need per-call evidence behind the spike.

​

Reading exec sentiment

Open Insights → Sentiment and lead with the net sentiment headline checked against the analyzed-message count next to it. The by-channel panel says where a dip lives; the resolution-lift tile says whether resolving still correlates with satisfaction. This is the page that fits on a slide; the other three are where you go when the slide raises a question.

​

Deep guides

Each surface has a full walkthrough:

• Read the Voice → Intelligence page — windows, keyword cloud, topic clusters, intents, categories, and operators.
• Read the Topic Intelligence dashboard — how topics are computed, the trend labels, and the API contract.
• Read the sentiment portfolio dashboard — net sentiment, the breakdowns, resolution lift, and a worked coaching example.
• Read the Voice Agent Quality Index (VAQI) — latency percentiles, barge-in, per-use-case thresholds, and a latency-regression walkthrough.

​

How this relates to Conversation Intelligence

The existing concept page Conversation Intelligence is not a fifth dashboard, and this page does not replace it. That page describes the scoring layer over your inbox and agent conversations (per-thread sentiment, topics, and quality score, tenant-defined operators, and the /stats/conversation-intelligence API): the pipeline that produces the labels the surfaces above roll up. Read it for how the signals are computed, stored, and queried; read this page to pick which console surface answers your question.