The deepest analytics layer for voice, chat, and email AI agents. Cozmo measures every conversation, surfaces what is working, and shows exactly where to invest next.
Every conversation, every workflow, every dollar: measured.
Most AI agent platforms give you a list of calls and call it analytics. Cozmo gives you a real analytics platform, built for the way multi-modal AI agents fail, succeed, and improve over time. Across voice, chat, and email, every chart drills down to the underlying conversations.
Cozmo AnalyticsLive
Conversations48,219
Completion87.3%
Conversion14.8%
Latency612ms
Conversation volume
VoiceChatEmail
Voice, chat, and email. Measured the same way.
Every channel is a first-class citizen, with the same filters, metrics, and drilldowns.
Voice
Voice analytics that understands how voice actually breaks.
End-to-end latency by pipeline stage, pickup rate, voicemail rate, interruptions, hang-up moments, caller country, language, time of day, and workflow.
Latency breakdownLive
StageP50P95Δ
STT first token142ms218ms0ms
Turn detection68ms112ms0ms
LLM first token284ms512ms0ms
TTS first byte96ms164ms0ms
Network42ms88ms0ms
End-to-end612ms1094ms0ms
Chat
Chat analytics that tracks every message and every thread.
Response speed, message engagement, resolved intents, human handoffs, sentiment shifts, and the exact messages that cause users to drop.
Chat funnelToday
Message sent12,840
Intent captured12,070
Action attempted10,400
Resolved9,502
Email
Email analytics treated as a peer, not an afterthought.
Open rate, reply rate, time-to-first-reply, thread depth, escalation, and outcome attribution for outbound and inbound email agents.
Email analytics30d rolling
Open rate42.1%
Reply rate18.4%
Time-to-reply38 min
Trial reactivation24.1%
Payment recovery19.8%
Onboarding drip18.4%
Feature announcement12.2%
Inside the Cozmo analytics platform.
Eight purpose-built layers, built to scale from your first 1,000 conversations to your hundred-millionth.
Conversation analytics
Every voice call, chat thread, and email exchange becomes a structured record the moment it ends. Cozmo extracts the transcript, sentiment arc, topics discussed, tools the agent called, data points captured, and the outcome. All queryable, all filterable, all exportable. Search by keyword, filter by sentiment, sort by cost, replay the audio, and tag conversations for training datasets in one click.
09:14· Voice · 3m 22sResolvedPositive$0.28
09:18· Chat · 1m 08sEscalatedNeutral$0.11
09:31· Email · asyncResolvedPositive$0.04
09:44· Voice · 6m 51sResolvedNegative$0.54
Workflow analytics
Every Cozmo workflow reports completion rate, failure reasons, step-level drop-off, retry health, and version regressions in real time. Compare workflow versions side-by-side, see which step lost the most users, and cluster failures by root cause automatically. When a workflow degrades, Cozmo alerts you before your customers notice.
Support v4.1
Healthy
Collections v2.8
Degraded
Lead qual v1.3
Healthy
Agent performance
Compare every agent on every metric: total conversations, completion rate, outcome success, average cost, average latency, CSAT, and error rate. Sort by anything. Drill into anything. The Agent Performance Matrix plots all your agents on a quality-versus-efficiency chart so you can see at a glance which ones are stars, which are workhorses, and which need attention.
High qualityLow qualityLow efficiencyHigh efficiency
Prospect funnel and conversion
For sales and lead-qualification agents, Cozmo tracks prospects from creation through every touch to conversion. See the full funnel: created, contacted, engaged, qualified, converted. Compare conversion by source, country, agent, and workflow. A world map shows where your prospects come from and where they convert best.
Created8,420
Contacted6,190
Engaged3,840
Qualified1,920
Converted740
Quality and safety monitoring
Incorrect outputs, guardrail trips, knowledge gaps, off-topic drift: Cozmo monitors every conversation and surfaces the moments that need human review. Each detected issue shows the agent's claim, the ground truth from your knowledge base, and a one-click action to flag or fix it. Knowledge gaps are auto-clustered: when 214 callers ask about refund timelines and your agent cannot answer, Cozmo groups them and drafts the KB article.
Knowledge gaps (auto-clustered)
Refund timelines214 callers
Account reinstatement98 callers
Late payment grace period71 callers
Latency and reliability
End-to-end latency broken down by stage, with P50, P95, and P99 percentiles, segmentable by agent, workflow, region, and provider. When a third-party provider degrades, Cozmo's provider health board shows it before your customers feel it. Failover events are logged. Concurrency is tracked against your capacity ceiling. Geographic latency is heat-mapped so you know when callers in one region are getting a worse experience than another.
Peak concurrency vs capacity
Cost and unit economics
Every dollar broken down by LLM, STT, TTS, telephony, tool calls, and platform: per agent, per workflow, per customer. Cost per conversation. Cost per outcome. Cost per qualified lead. Cost per booked meeting. Cozmo forecasts your end-of-month spend, alerts you when a workflow's costs spike, and compares your current providers against alternatives.
Voice$0.31
Chat$0.12
Email$0.04
LLM cost up 38% Tuesday — investigate
Full API access
Every metric on this page is queryable via REST or SQL. Stream events to your warehouse, embed dashboards in your own product, pipe alerts to Slack, PagerDuty, or your CRM. Webhooks fire on every conversation, every workflow run, every threshold breach. Direct sync to Snowflake, BigQuery, Redshift, and Databricks. Real-time event streams via webhooks and Kafka.
// GET conversations with filters
GET /v1/conversations?
modality=voice
&sentiment=negative
&outcome=escalated
&from=2026-05-01
// → returns 284 conversations
Ready to see Cozmo AIhandle real operations?
We’ll open up a real workflow and show you exactly what happens with voice, documents, decisions and system updates.
