Documentation Index
Fetch the complete documentation index at: https://koreai.mintlify.app/llms.txt
Use this file to discover all available pages before exploring further.
Insights provides a comprehensive analytics and monitoring layer for your AI agent program — significantly expanding the depth of visibility available in previous platform generations. The section includes pre-built executive dashboards, specialized analytics pages, voice channel diagnostics, and a configurable pipeline framework that lets you define your own evaluation metrics alongside the built-in ones.
Overview
| Page | Purpose |
|---|
| Dashboard | Pre-built executive dashboards with KPIs, trend charts, ROI metrics, and conversation data |
| Analytics | Event volume, LLM performance, session and trace exploration with granular time ranges |
| Billing and Usage | Published billing-unit usage reporting |
| Agent Performance | Per-agent quality scores across evaluation dimensions |
| Quality Monitor | Quality health tracking with hallucination, safety, and context scores |
| Customer Insights | Intent distribution, sentiment trajectory, and frustration analysis |
| Feedback | End-user feedback captured from chat sessions, with rating and comment filters |
| Voice Analytics | Call quality, ASR accuracy, end-to-end latency, and barge-in metrics |
| Agent Transfer | Efficiency and performance metrics for human-handed conversations |
| Pipelines | Pre-built and custom analytics pipelines with a free-flow editor for your own evaluation logic |
Before You Begin
Confirm the following before working with Insights pages:
- You must have at least Viewer-level access to the project.
- Agent Performance, Quality Monitor, and Customer Insights require analytics pipelines enabled in Settings. Without pipelines, these pages show a placeholder.
- Voice Analytics requires at least one voice channel deployment to generate data.
Accessing Insights
Navigation: Project → Sidebar → Insights
Dashboard
The Dashboard page provides a pre-built executive overview of your AI agent program. It displays key performance indicators, trend charts, and conversation data for the selected time period, giving stakeholders immediate visibility without any configuration.
Navigation: Project → Insights → Dashboard
Date range selector
Use the toggle in the top-right corner to select 7d, 30d (default), or 90d. Changing the date range refreshes all dashboard data.
KPI Metric Cards
| Metric | Description |
|---|
| Conversations | Total conversation count. |
| Containment Rate | Percentage of sessions resolved without human escalation. Warning icon appears if low. |
| Quality Score | Aggregated quality score across evaluated conversations. Dash if unavailable. |
| Avg Sentiment | Average sentiment score across all conversations. Dash if insufficient data. |
| Cost Savings | Estimated cost savings versus human-handled conversations. Negative indicates no cost parity. |
| Escalation Rate | Percentage of sessions requiring human escalation. |
| Tab | What it shows |
|---|
| Overview | Volume and containment trend chart, and cost breakdown. |
| Trends | Longitudinal trends for key metrics. |
| ROI | Return on investment metrics comparing agent costs to human-handled baselines. |
| Conversations | Filterable list of individual conversations with status and outcome details. |
Analytics
The Analytics page monitors event volume, LLM performance, and session metrics.
Navigation: Project → Insights → Analytics
Time range controls
Analytics supports granular time ranges: 30m, 1h, 3h, 6h, 12h, 24h, 2d, 7d, 30d, or a Custom range with specific start and end timestamps.
Overview Tab
| Metric | Description |
|---|
| Sessions | Total sessions in the selected period. |
| Messages | Total messages exchanged. |
| LLM Calls | Total LLM API calls made by agents. |
| Errors | Total errors encountered during agent execution. |
| Tokens | Total LLM tokens consumed. |
| Cost | Estimated cost based on token usage and model pricing. |
| Tab | Purpose |
|---|
| LLM Performance | Model-level metrics including latency, token usage per call, and error rates. |
| Sessions Explorer | Browse and filter individual sessions with conversation details and traces. |
| Traces Explorer | Search and inspect trace events across sessions for debugging. |
| Query | Run custom analytics queries against project event data. |
Billing and Usage
The Billing and Usage page displays published billing-unit usage reporting for your project.
Navigation: Project → Insights → Billing and Usage
Use the time range selector to view usage for the last 7 days, 30 days, or 90 days. Billing data appears after materialized batches apply to the reporting plane.
The Agent Performance page monitors and compares agent quality across all evaluation dimensions. Use the date range selector (7d, 30d, or 90d) to adjust the reporting period.
Navigation: Project → Insights → Agent Performance
This page requires analytics pipelines. Enable analytics pipelines in Settings to start tracking agent quality, hallucination rates, knowledge gaps, and more.
Quality Monitor
The Quality Monitor page tracks quality health across all evaluation dimensions.
Navigation: Project → Insights → Quality Monitor
Quality Health summary
A banner at the top displays the number of evaluated conversations, the aggregated quality score, and counts of critical and healthy dimensions.
Evaluation dimensions
| Dimension | Description | Target |
|---|
| Overall Quality | Aggregated quality score across all dimensions. | Higher is better |
| Hallucination Rate | Percentage of responses with unsupported claims or inaccuracies. | Lower is better |
| Knowledge Gaps | Percentage of queries where the agent lacked sufficient knowledge. | Lower is better |
| Safety Score | Percentage of responses passing guardrail safety checks. | Guardrail pass |
| Context Preservation | Percentage of responses maintaining correct conversational context. | Higher is better |
Customer Insights
The Customer Insights page helps you understand customer queries and sentiment.
Navigation: Project → Insights → Customer Insights
KPI Metric Cards
| Metric | Description |
|---|
| Total Conversations | Total conversations analyzed in the selected period. |
| Unique Intents | Number of distinct intents identified. |
| Avg Sentiment | Average sentiment score across all conversations. |
| Frustration Rate | Percentage of conversations where the system detected user frustration. |
| Resolution Rate | Percentage of conversations that reached successful resolution. |
Feedback
The Feedback page surfaces end-user feedback captured from chat sessions in your project. Use it to review ratings, read verbatim comments, and filter by agent or channel to understand how users perceive the agent experience.
Navigation: Project → Insights → Feedback
Date range selector
Use the toggle at the top to select 7d, 30d (default), or 90d. Changing the date range refreshes all feedback data.
Filters
| Filter | Description |
|---|
| All ratings | Filter by feedback rating. Use the dropdown to select a specific rating or view all ratings. |
| Comment: any | Filter by comment presence. Choose whether to show all feedback, only entries with comments, or only entries without comments. |
| Agent name | Filter by the agent that handled the conversation. Type an agent name to narrow results. |
| Channel | Filter by the channel through which the conversation occurred. Type a channel name to narrow results. |
Voice Analytics
The Voice Analytics page provides a dedicated dashboard for monitoring call quality, speech recognition accuracy, and end-to-end latency across the voice processing pipeline. Use the date range selector (24h, 7d, or 30d) to adjust the reporting period.
Navigation: Project → Insights → Voice Analytics
KPI Metric Cards
| Metric | Description |
|---|
| Total Calls | Number of voice calls in the selected period. |
| Avg MOS | Average Mean Opinion Score for call quality (scale 1–5). |
| ASR Quality | Automatic Speech Recognition quality score (0–100, higher is better). |
| E2E Latency | End-to-end latency in milliseconds for the voice processing pipeline. Covers the full round-trip from user speech input through ASR, LLM processing, and TTS output. |
| Barge-In Rate | Percentage of calls where the user interrupted the agent mid-response. |
| DTMF Fallback | Percentage of calls that fell back to touch-tone input. |
| Chart | Description |
|---|
| Network Quality and Call Volume | MOS scores and call count trends over the selected period. Use this to correlate call quality dips with volume spikes. |
| Speech Recognition Quality (ASR) | ASR quality scores over time. Monitor for degradation that may indicate noisy environments or model drift. |
Track E2E Latency trends after model or pipeline changes. Even small latency increases can affect caller experience and barge-in rates.
Agent Transfer
The Agent Transfer page, titled Queues and Agents, displays efficiency and performance metrics for conversations handed off to human operators.
Navigation: Project → Insights → Agent Transfer
Use the date range selector (Today, 7d, or 30d) to adjust the reporting period. The page organizes data into three sections:
| Section | What it shows |
|---|
| Efficiency | Transfer efficiency metrics split by Voice, Chat, and Transfers. |
| Queue Performance | Queue-level metrics, including wait times and handling rates. |
| Agent Performance | Human agent performance metrics for transferred conversations. |
Pipelines
The Pipelines page is where Insights shifts from pre-built dashboards to customer-defined analytics. While the platform ships with a comprehensive set of built-in pipelines, the real power lies in the ability to create your own — defining exactly what to evaluate, when to trigger evaluation, and which metrics to surface.
Navigation: Project → Insights → Pipelines
Each pipeline card displays its enabled or disabled status, trigger count, and last processed timestamp. Enable pipelines to start populating data in Agent Performance, Quality Monitor, and Customer Insights.
Tabs
| Tab | Purpose |
|---|
| Built-in | Pre-configured pipelines that ship with the platform, ready to enable. |
| Custom | User-defined processing workflows created using the free-flow editor. |
| Recent Runs | Pipeline execution history with timestamps, durations, and status. |
| Data | Pipeline output data available for dashboard integration and export. |
Built-in Pipelines
The platform ships with six pre-built pipelines covering the most common evaluation needs. Enable each with a single toggle:
| Pipeline | Description |
|---|
| Sentiment Analysis | Per-message sentiment scoring with trajectory analysis. |
| Intent Classification | Classifies conversation intent using LLM analysis. |
| Quality Evaluation | LLM-as-judge quality evaluation with configurable rubric. |
| Hallucination Detection | Detects unsupported claims, contradictions, and factual inaccuracies. |
| Knowledge Gap Analysis | Identifies gaps in knowledge base coverage. |
| Guardrail Analysis | Evaluates guardrail effectiveness, detecting false positives and negatives. |
Custom Pipelines
Custom pipelines let you define your own analytics logic using a free-flow visual editor. Use this mechanism to build organization-specific evaluation criteria that go beyond the built-in set.
How the editor works
The free-flow editor presents a visual canvas where you define a pipeline as a sequence of connected steps. The core pattern: when a trigger fires, run one or more evaluation steps, then produce metrics that feed into dashboards.
Pipeline structure
| Component | Description |
|---|
| Trigger | Defines when the pipeline runs. Triggers can fire on every conversation, on a schedule, on specific events, or on a filtered subset of sessions. |
| Evaluation steps | The processing logic applied to each triggered conversation. Steps can call LLMs, apply regex rules, check knowledge base coverage, compute scores, or run custom code. |
| Metrics output | The results the pipeline produces. Metrics are named, typed values (counts, percentages, scores) that appear in the Data tab and can attach to dashboards. |
Start with the built-in pipelines to establish baselines, then create custom pipelines for organization-specific quality dimensions — for example, regulatory compliance checks, brand voice adherence, or domain-specific accuracy.
Working with Date Ranges
All Insights pages respect the selected date range. Use 7 days for operational monitoring and quick health checks. Use 30 days for monthly reviews and reporting. Use 90 days for trend analysis and strategic planning.
Compare 30-day periods before and after agent changes to measure the impact of your improvements.
