Knowledge Base Academy Community My Tickets
Fabric 8.4
Fabric 8.4

Observation Dashboard

The Observation dashboard is the primary interface for monitoring AI agent activity in production. It provides a consolidated view of conversation quality, volume, and risk across a configurable time window, combining high-level KPI widgets with a filterable conversation table that links directly to individual session reviews.

Observation dashboard

Accessing the Dashboard

Navigate to:

AI Data Fusion > Observation

The dashboard loads with the Last 7 Days time window selected by default.

Dashboard Layout

The dashboard is organized in four areas, top to bottom:

  1. Header bar - page title, inline summary stats for the selected period, and the duration filter
  2. Info cards - four KPI cards showing averages for the selected period
  3. Chart widgets - charts covering topics, sentiment, volume, satisfaction, resolution, and cost. This is the default widgets, which can be changed, replaced, reorganized, as well as extended, per needs
  4. Conversation table - a filterable list of individual sessions with a summary per row

Header Bar

The header bar displays the title Conversation Analytics alongside a set of inline summary indicators that give an immediate at-a-glance view of the selected period:

  • Total - total number of conversations
  • Not Resolved % - percentage of conversations tagged as not resolved
  • Follow-ups - count of conversations requiring follow-up
  • At Risk % - percentage of conversations flagged as at risk
  • Resolved % - percentage of conversations tagged as resolved

The duration filter is on the right side of the header bar.

Duration Filter

The duration filter controls the time window for all widgets and the conversation table simultaneously.

Option Use case
Last 24 Hours Monitoring activity since the last deployment or investigating a recent spike
Last 7 Days Default weekly review; identifies emerging patterns
Last 30 Days Monthly performance assessment and trend tracking
Last 90 Days Quarterly analysis and long-term quality trends

Changing the duration refreshes all widgets and the conversation table without reloading the page.

Info Card Widgets

Four info cards are displayed side by side, each showing an average for the selected period:

  • Avg. Resolution Rate - percentage of conversations that reached a resolved state
  • Avg. Tokens / Conversation - mean total tokens per session, with an input/output breakdown shown below the main figure
  • Avg. Steps / Conversation - mean number of agent execution steps per session; indicates conversation complexity
  • Avg. Response Time - mean end-to-end response time per turn, in seconds

Each card's query is configurable. See Configuring Observation for how to add or modify widgets.

Chart Widgets

Chart widgets are displayed in a 2 or 3 grid (depending on width and screen resolution), covering the key dimensions of conversation quality and cost.

Below are the default widgets:

Top Conversation Topics - Bar chart ranking topics by session count. Reveals which subjects drive the most volume (e.g., Billing Inquiry, Account Update, Technical Support).

Customer Sentiment - Pie chart showing the split between Positive, Neutral, and Negative conversations. Gives an immediate read on overall customer experience.

Conversation Volume Over Time - Line chart of daily session count. Highlights traffic patterns, spikes after deployments, and quiet periods.

Satisfaction Trend (%) - Line chart of satisfaction percentage over time. Use this to track whether quality is improving or declining following agent changes.

Resolution Pathway - Bar chart showing how far conversations progress through the resolution stages (Initial Contact → Issue Identified → Solution Provided → Customer Engagement → Resolution Achieved). Identifies where conversations tend to drop off.

Token Consumption - Bar chart of input, cache and output token usage over time. Use for cost tracking and detecting unusual consumption spikes.

All widgets are configurable. See Configuring Observation for how to add, modify, or replace widgets.

Conversation Table

The Conversation Details table lists individual sessions for the selected period. A filter dropdown above the table narrows the list by conversation type.

Columns

Column Data field Description
Date CREATION_DATE Session timestamp, formatted as YYYY-MM-DD HH:MM
Customer CUSTOMER_ID Customer identifier (IID), displayed with a # prefix
Topics TOPICS Auto-tagged topic labels; up to two topics shown, with hover text for the full list
Resolution RESOLUTION_STATUS Colored badge reflecting the LLM-assigned resolution signal:
  • Resolved — green; conversation reached a successful conclusion
  • Pending — amber; issue was not fully resolved
  • Escalated — red; conversation was escalated
  • Unknown — gray; session has not yet been analyzed
Session lifecycle states (OPEN, CLOSED, FAILED) may also appear if the session record is surfaced directly.
Risk RISK_LEVEL Colored badge: Low (green), Medium (orange), High (red)
Avg Response AVG_RESPONSE_TIME Mean response time per agent turn in this session, shown in seconds (e.g., 5.2s)
Submitted By SUBMITTED_BY User or system identity that initiated the session
App ID APP_ID The AI Fusion application that handled the session
Summary CHAT_SUMMARY LLM-generated summary snippet; hover for full text

Filter Modes

Filter What it shows
All All conversations in the selected period (default)
High Risk Conversations Sessions with HIGH risk level
Unresolved Conversations Sessions tagged as Pending or Escalated
Customer ID All sessions for a specific customer IID. Selecting this option reveals a text input for the customer ID
Submitted By All sessions initiated by a specific user, for example a CRM representative. Selecting this option reveals a text input
App ID All sessions handled by a specific application. Selecting this option reveals a text input

Exporting Data

Click the Export List button above the conversation table to download the current view (respecting the active duration and filter) as a file. Use this for reporting, sharing with stakeholders, or loading into external analytics tools.

Click any row in the conversation table to open the Conversation Review page for that session. The review page shows the full message history, auto-tags, user feedback, and execution trace.

Next article: Conversation Review

Observation Dashboard

The Observation dashboard is the primary interface for monitoring AI agent activity in production. It provides a consolidated view of conversation quality, volume, and risk across a configurable time window, combining high-level KPI widgets with a filterable conversation table that links directly to individual session reviews.

Observation dashboard

Accessing the Dashboard

Navigate to:

AI Data Fusion > Observation

The dashboard loads with the Last 7 Days time window selected by default.

Dashboard Layout

The dashboard is organized in four areas, top to bottom:

  1. Header bar - page title, inline summary stats for the selected period, and the duration filter
  2. Info cards - four KPI cards showing averages for the selected period
  3. Chart widgets - charts covering topics, sentiment, volume, satisfaction, resolution, and cost. This is the default widgets, which can be changed, replaced, reorganized, as well as extended, per needs
  4. Conversation table - a filterable list of individual sessions with a summary per row

Header Bar

The header bar displays the title Conversation Analytics alongside a set of inline summary indicators that give an immediate at-a-glance view of the selected period:

  • Total - total number of conversations
  • Not Resolved % - percentage of conversations tagged as not resolved
  • Follow-ups - count of conversations requiring follow-up
  • At Risk % - percentage of conversations flagged as at risk
  • Resolved % - percentage of conversations tagged as resolved

The duration filter is on the right side of the header bar.

Duration Filter

The duration filter controls the time window for all widgets and the conversation table simultaneously.

Option Use case
Last 24 Hours Monitoring activity since the last deployment or investigating a recent spike
Last 7 Days Default weekly review; identifies emerging patterns
Last 30 Days Monthly performance assessment and trend tracking
Last 90 Days Quarterly analysis and long-term quality trends

Changing the duration refreshes all widgets and the conversation table without reloading the page.

Info Card Widgets

Four info cards are displayed side by side, each showing an average for the selected period:

  • Avg. Resolution Rate - percentage of conversations that reached a resolved state
  • Avg. Tokens / Conversation - mean total tokens per session, with an input/output breakdown shown below the main figure
  • Avg. Steps / Conversation - mean number of agent execution steps per session; indicates conversation complexity
  • Avg. Response Time - mean end-to-end response time per turn, in seconds

Each card's query is configurable. See Configuring Observation for how to add or modify widgets.

Chart Widgets

Chart widgets are displayed in a 2 or 3 grid (depending on width and screen resolution), covering the key dimensions of conversation quality and cost.

Below are the default widgets:

Top Conversation Topics - Bar chart ranking topics by session count. Reveals which subjects drive the most volume (e.g., Billing Inquiry, Account Update, Technical Support).

Customer Sentiment - Pie chart showing the split between Positive, Neutral, and Negative conversations. Gives an immediate read on overall customer experience.

Conversation Volume Over Time - Line chart of daily session count. Highlights traffic patterns, spikes after deployments, and quiet periods.

Satisfaction Trend (%) - Line chart of satisfaction percentage over time. Use this to track whether quality is improving or declining following agent changes.

Resolution Pathway - Bar chart showing how far conversations progress through the resolution stages (Initial Contact → Issue Identified → Solution Provided → Customer Engagement → Resolution Achieved). Identifies where conversations tend to drop off.

Token Consumption - Bar chart of input, cache and output token usage over time. Use for cost tracking and detecting unusual consumption spikes.

All widgets are configurable. See Configuring Observation for how to add, modify, or replace widgets.

Conversation Table

The Conversation Details table lists individual sessions for the selected period. A filter dropdown above the table narrows the list by conversation type.

Columns

Column Data field Description
Date CREATION_DATE Session timestamp, formatted as YYYY-MM-DD HH:MM
Customer CUSTOMER_ID Customer identifier (IID), displayed with a # prefix
Topics TOPICS Auto-tagged topic labels; up to two topics shown, with hover text for the full list
Resolution RESOLUTION_STATUS Colored badge reflecting the LLM-assigned resolution signal:
  • Resolved — green; conversation reached a successful conclusion
  • Pending — amber; issue was not fully resolved
  • Escalated — red; conversation was escalated
  • Unknown — gray; session has not yet been analyzed
Session lifecycle states (OPEN, CLOSED, FAILED) may also appear if the session record is surfaced directly.
Risk RISK_LEVEL Colored badge: Low (green), Medium (orange), High (red)
Avg Response AVG_RESPONSE_TIME Mean response time per agent turn in this session, shown in seconds (e.g., 5.2s)
Submitted By SUBMITTED_BY User or system identity that initiated the session
App ID APP_ID The AI Fusion application that handled the session
Summary CHAT_SUMMARY LLM-generated summary snippet; hover for full text

Filter Modes

Filter What it shows
All All conversations in the selected period (default)
High Risk Conversations Sessions with HIGH risk level
Unresolved Conversations Sessions tagged as Pending or Escalated
Customer ID All sessions for a specific customer IID. Selecting this option reveals a text input for the customer ID
Submitted By All sessions initiated by a specific user, for example a CRM representative. Selecting this option reveals a text input
App ID All sessions handled by a specific application. Selecting this option reveals a text input

Exporting Data

Click the Export List button above the conversation table to download the current view (respecting the active duration and filter) as a file. Use this for reporting, sharing with stakeholders, or loading into external analytics tools.

Click any row in the conversation table to open the Conversation Review page for that session. The review page shows the full message history, auto-tags, user feedback, and execution trace.

Next article: Conversation Review

© Copyright 2026 K2VIEW