Conversation Mining - Kore.ai Docs

Conversation Mining provides supervisors with a centralized workspace to review, analyze, monitor, and audit customer interactions processed by Quality AI. It supports AI Agent, Human Agent, and AI Agent → Human Agent handoff conversations, including processed and pending interactions. For AI Agent → Human Agent handoff interactions, the system represents the AI Agent and Human Agent portions as separate evaluation records. Supervisors can filter, review, audit, and evaluate each record independently. The Handled By field identifies whether a record belongs to an AI Agent or Human Agent segment. Conversation Mining brings together conversation quality, evaluation scores, customer experience, business metadata, sentiment, topics, intents, emotions, resolution outcomes, and operational insights. These insights help supervisors conduct audits, coach agents, investigate issues, and analyze performance across Conversation Mining, AI-Assisted Audit, and Agent Dashboards.

Use Contact Direction together with Queue, Experience Flow, Channel, Evaluation Form, and other filters to analyze inbound and outbound interactions.

Contact Direction

Use Contact Direction together with Queue, Experience Flow, Channel, Evaluation Form, and other filters to distinguish between Inbound and Outbound interactions. Contact Direction applies to both Human Agent and AI Agent conversations. The available results must respect the selected direction when combined with routing, channel, agent, and evaluation filters.

Interactions Page

The Interactions page includes two tabs:

Processed: Displays conversations that have completed Quality AI processing and are available for review, audit, and evaluation.
Pending: Displays conversations awaiting processing or that failed validation. Successfully processed conversations move to the Processed tab.

When you enable Agent Accept & Dispute, the Processed tab displays evaluation review status in the Status column and Audit Status filter.

Key Capabilities

Review processed conversations for coaching, investigations, and quality analysis.
Analyze evaluation scores, sentiment, topics, intents, emotions, and resolution outcomes.
Review AI Agent, Human Agent, and AI-to-Human handoff conversations independently.
Open the corresponding AI or Human audit segment directly from each conversation record.
Identify processing failures independently for versioned AutoQA features, including By Question V1/V2 and By Agent V1/V2, when feature-level processing details are available.
Filter conversations using queues, experience flows, agents, intents, keywords, channels, contact direction, evaluation metrics, and custom fields.
Apply experience filters such as L1, L2, L3 intents, generated intents, resolution status, and duration status.
Customize workspace with configurable columns, saved views, and timeline navigation.
Track review status including agent acknowledgement and dispute status for evaluated conversations.
Maintain V1 and V2 processing statuses independently. A failure in one version doesn’t affect the other.

Conversation Records

The conversation type determines the number of rows displayed.

Conversation Type	Records Displayed
Human-only Conversation	One Human Agent record.
AI-only Conversation	One AI Agent record.
AI Agent → Human Agent Handoff	Two separate records (AI Agent and Human Agent segments).

Each record opens its corresponding audit segment.

AI Agent → Human Agent Handoff

When an interaction is transferred from an AI Agent to a Human Agent, Conversation Mining creates two separate evaluation records:

AI Agent Record: Represents the AI Agent segment and uses its assigned Experience Flow.
Human Agent Record: Represents the Human Agent segment and uses its assigned Queue.

Each segment is evaluated independently. Scores, audit results, metrics, moments, intents, and other segment-specific data remain separate between the two records.

Record Behavior

Each evaluation segment appears as an independent record.
Selecting a Human Agent record opens the Human Agent audit segment.
Selecting an AI Agent record opens the AI Agent audit segment.
AI Agent records use Experience Flow assignments.
Human Agent records use Queue assignments.

Processed Tab

The Processed tab (Processed Conversations) displays conversations that have completed Quality AI processing. Select a conversation to open the AI-Assisted Audit screen, where you can review the transcript, evaluation results, sentiment, audit details, and other conversation insights. For AI-to-Human conversations:

Selecting the AI Agent row opens the AI Agent audit segment.
Selecting the Human Agent row opens the Human Agent audit segment.

Processing Status

Indicates the processing status of the conversation and applicable Quality AI features. For AutoQA, processing status is maintained independently for By Question V1, By Question V2, By Agent V1, and By Agent V2, when configured. Each feature version retains its own processing result. A failure in one version doesn’t overwrite, mask, or change the processing result of another version. For example, a conversation may have a successful By Question V1 result while By Question V2 has failed. The UI must preserve and display these version-specific results independently.

The Pending tab is available only when Automation AI, Agent AI, or Contact Center AI (CCAI) conversation sources are enabled for the workspace.

Access Process Tab Conversations

Navigate to Quality AI > AutoQA > Conversation Mining > Processing.

Filter by Processing Status

Supervisors use the Status filter on the Pending tab to view conversations by processing state. The filter includes the following options:

Select All: Displays all pending conversations regardless of status.
Pending: Displays conversations queued for processing.
In Progress: Displays conversations currently being processed.
Failed: Displays conversations that failed validation. Selecting this option also reveals failure-specific sub-categories.

Conversation Grid Details

The Conversation Mining grid displays one record for every evaluation segment.

Column	Description
Handled by	Displays whether an AI Agent or Human Agent handles the interaction
Assigned to	Displays the routing assignment for the interaction as a colored tag. For Human Agent conversations, this displays the assigned Queue. For AI Agent conversations, this displays the assigned Experience Flow. The applicable Queue or Experience Flow determines the evaluation form used for scoring.
Agents	Displays the name of the agent who last handled and ended the interaction.
Processing Status	Displays the Quality AI processing state and applicable feature-level processing results, including version-specific AutoQA processing statuses. The system maintains AutoQA feature statuses independently for By Question V1, By Question V2, By Agent V1, and By Agent V2, enabling it to identify failures separately by metric version. Conversation Intelligence generates outputs such as sentiment, topics, intents, and taxonomy-based insights.
Actions	Displays available actions such as Bookmark and Audit.
Flow or Queue	Displays the assigned Experience Flow for AI Agent conversations or Queue for Human Agent conversations.
Kore Evaluation Score	Displays the AutoQA evaluation score, when available.
Sentiment Score	Displays the customer sentiment score derived from the conversation.
Supervisor Audit Score	Displays the supervisor-assigned score for audited or manually evaluated interactions and highlights manual evaluation metrics (fatal metrics count) with a consistent visual indicator.
Start Time	Displays the conversation start time (for example, 24 May 2024, 1:17:10 PM).
Duration	Displays the interaction duration.
Moments	Displays detected conversation moments, including Questions Adherence (met metrics), Violations (speech-based issues), and Omissions (unmet metrics such as playbook steps, dialog tasks, and question-level checks). Select an interaction to open the conversation history and view Moment metrics on the AI-Assisted Manual Audit page.
Sentiment Trend	Displays the overall sentiment progression.
Emotions	Displays customer emotional states or tracked emotions during the conversation.
Conversation ID	Displays the user-defined identifier for custom tracking.
Custom Conversation ID	Displays the unique system-generated identifier used as the primary key.
Configured Intents	Displays mapped intents across L1, L2, and L3 with sentiment indicators. A color-coded map (red, green, and gray) indicates the sentiment of each detected topic.
Overall Resolution	Displays the resolution outcome, such as Resolved or Unresolved, based on whether the conversation addresses all L3 intents.
Generated Intents	Displays AI-generated intents for each conversation. Displays No form assigned when the system can’t evaluate the conversation because no applicable direction-specific evaluation form is assigned or available.
Direction	Displays the channel and contact direction, for example, Voice - Inbound or Chat - Inbound.
Status	Displays the evaluation review or audit allocation status of the interaction. When Agent Accept & Dispute is enabled, this includes the applicable acknowledgement and dispute status.

You must create bookmarks in Settings before tagging interactions for future reference.

AI Agent Conversation Behavior

AI Agent conversation segments differ from Human Agent segments. For AI Agent conversations:

Experience Flow replaces Queue.
AI evaluation forms are used.
AI-specific evaluation metrics are used.

Speech-related metrics that require Human Agent speech analysis aren’t generated for AI Agent conversations. This includes: Empathy Score, Dead Air, Cross-talk, Speaking Rate, Hold, and Crutch Word Score. These metrics are applicable only to Human Agent conversations. AI Agent conversations mustn’t display or return results for these Human Agent-only metrics.

Pending Tab

The Pending tab displays conversations that haven’t completed Quality AI processing. This includes conversations in Pending, In Progress, or Failed processing states.

Pending: The conversation is queued and awaiting processing.
In Progress: The conversation is currently being processed.
Failed: The system doesn’t process the conversation successfully because it failed validation or processing. When available, the system displays the failure reason and applicable failure category.

Conversations that complete processing successfully are moved to the Processed tab.

The Pending tab is available only when Automation AI, Agent AI, or Contact Center AI (CCAI) conversation sources are enabled for the workspace.

Access Pending Conversations

Navigate to Quality AI > AutoQA > Conversation Mining > Pending.

Pending Tab Columns

Column	Description
Handled By Agent	Displays the name of the agent who handled the pending conversation.
Conversation ID	Displays the conversation identifier for the interaction.
Start Date	Displays the date when the conversation started.
End Date	Displays the date when the conversation ended.
Processing Status	Displays the current state of conversation processing.

Filter By Processing Status Values

Use the Status filter on the Pending tab to filter conversations by processing state.

Status or Filter Option	Description
Select All	Displays all pending conversations.
Pending	Displays conversations queued for processing. If Scheduled Processing is enabled, conversations remain queued until the next processing window.
In Progress	Displays conversations that are currently being processed.
Failed	Indicates that the conversation failed validation and isn’t processed. Quality AI records the failure reason and may provide failure-specific categories.

Filter Conversations

Use the toolbar at the top of the Conversation Mining page to locate and manage conversations. Filters are available from the Conversation Mining toolbar and apply to the currently selected Processed or Pending tab. Filter availability and behavior may vary depending on the selected tab and the type of filter.

Control	Description
Columns	Include or exclude specific fields in the Interactions dashboard.
Bookmarks	Assign interactions to bookmarks and view all bookmarks associated with an interaction.
Date and Time Range	Select the time range for conversations. The default range is the last 7 days.
Filters	Open the filter panel to refine the displayed conversations.

Add New Filter

Select Filters from the Conversation Mining toolbar.
Select + Add New Filter.
Select a filter category.
Configure the filter and apply it.

The selected filters apply to the active Processed or Pending tab.

Filter Categories

The filter panel includes:

Filter by Efficiency

Filter by Efficiency analyzes emotional, behavioral, and qualitative aspects of customer interactions using sentiment, interaction metadata, and operational filters.

Handled By

Filter conversations based on the type of agent that handled the interaction:

AI Agent: Shows conversations handled by an AI Agent.
Human Agent: Shows conversations handled by a Human Agent.

Channels

Use this filter to view conversations based on how the interaction took place and its direction. Chat:

Inbound: Shows chat conversations initiated by the customer.
Outbound: Shows chat conversations initiated by an agent or the organization.

Voice:

Inbound: Shows voice conversations initiated by the customer, such as incoming calls to an agent or AI Agent.
Outbound: Shows voice conversations initiated by an agent or the organization, such as outgoing calls to a customer.

Audit Status

Filter conversations based on their audit status:

Assigned: Conversations assigned for audit.
Audited: Conversations that have completed an audit.
Not Assigned: Conversations that aren’t assigned for audit.
Resolved: Conversations with a resolved audit status.
Closed: Conversations with a closed audit status.
Pending Agent Review: Conversations awaiting agent review.
Pending Supervisor Review: Conversations awaiting supervisor review.

Audit Allocations

Select Audit Allocations to filter and view interactions associated with specific audit batches or allocation groups directly from Conversation Mining without manually filtering individual Conversation IDs.

Duration Status

Select a Duration Status to control which interactions appear based on evaluation eligibility and duration:

Option	Description
All (default)	Displays all interactions regardless of duration.
Evaluated	Displays interactions whose duration meets the configured evaluation threshold and for which an evaluation score is generated.
Below Threshold	Displays interactions whose duration is below the configured evaluation threshold and therefore are excluded from scoring.
Duration Unavailable	Displays interactions where the required duration information is missing or the system can’t determine.

Session ID

Enter a Session ID to find a specific conversation session.

Conversation ID

Enter a Conversation ID to find a specific conversation.

Custom Conversation ID

Enter a Custom Conversation ID to find a conversation using an organization-specific identifier.

Handled AI Agent

Filter conversations handled by a specific AI Agent.

Experience Flow

Select or search for the Experience Flow associated with an AI-handled conversation.

Human Agent

Filter conversations handled by a Human Agent.

Queues

Select one or more Queues associated with human-handled conversations.

Agents

Select one or more Agents to filter conversations handled by specific human agents.

Combined AI Agent and Human Agent Filters

When filters for both AI Agent and Human Agent are applied, Conversation Mining uses OR logic between the two agent types. Results include conversations that match either the AI Agent criteria or the Human Agent criteria. Filters within each agent type continue to follow their configured filter behavior.

Advanced Performance Filters

Enable additional performance-based filters.

Filter	Description
Average Handling Time (AHT)	Filter using the minimum or maximum handling time (in seconds).
Deviation from AHT	Filter interactions based on deviation from the average handling time.
Deviation %	Filter interactions by the percentage deviation from AHT.
Number of Transfers	Filter interactions by the number of transfers.

Filter by Experience

This category helps analyze emotional and qualitative aspects of customer interactions using sentiment and emotion-based filters.

Filter	Description
Avg. Waiting Time	Select the average waiting time, in seconds, for the selected date range.
Sentiment Trend	Analyzes conversation sentiment across different stages using Opening and Closing sentiment options such as Positive, Neutral, Negative, or Any to evaluate customer experience and conversation outcomes.
Emotions	Helps identify interactions based on detected emotions such as Anger, Fear, Confusion, Sadness, and Happy for targeted emotional and behavioral analysis.
Sentiment Score	Set the minimum and maximum sentiment range for positive (higher) or negative (lower) sentiment.
CSAT	Filter by customer satisfaction score to view and drill down into interaction distribution across score ranges based on feedback responses.
Configured Intents	Filter by topic or intent across L1, L2, L3. Child lists update based on parent selection.
Generated Intents	Select the AI-generated or configured intents based on your intents. Select Configured Intent levels in order: L1 first, then L2, then L3. You must select at least one level to apply the filter.
Overall Resolution	Filter by Resolved or Unresolved (both selected by default). The system marks a conversation as resolved when it defines all applicable L3 intents.

You can view the CSAT and Avg. Waiting Time filter parameters only when the Connector is in disabled mode.

Filter by Behavior

This category analyzes filters based on agent conduct and interaction dynamics.

Filter	Description
Metric Name	Enables you to select a single evaluation metric (shown as a tag) and filter interactions by Pass or Fail. Manual metrics are visually indicated for audit identification.
Metric Qualification	Filter interactions by metric outcome. When you select a Dynamic By Question metric in the Metric Name filter, the system enables the following options: - Pass: Trigger detected, and the response matched the expected utterance (Adhered). - Fail: Trigger detected, but the response didn’t match the expected utterance (Non-Adhered).
Language	Select a conversation language to add from the dropdown list.
Empathy Score (Human Agent only)	Measures how well the agent responds to the customer’s emotional state. A higher score indicates a more empathetic interaction.
Crutch Word Score (Human Agent only)	Measures the use of filler words (for example, `um`, `uh`) by the agent. A higher score indicates greater use of crutch words.
Agent Playbook Adherence	Displays the Agent AI playbook adherence percentage for an interaction.
Kore Evaluation Score	Displays the automated QA score for an interaction, based on the evaluation form assigned to an interaction’s queue.
Fatal Interactions	The system displays interactions that contain fatal issues and excludes them from evaluation scoring.
Validation Failure	Displays interactions that fail validation and aren’t processed for evaluation. You can use Form, Channel, and Language to refine results.

Agent Playbook Adherence is available only when a Playbook is configured for the selected form.

Filter by Keywords

This enables you to search, filter, and review conversations using specific keywords or phrases. You can filter by included or excluded keywords and by speaker, with results linked directly to the QA audit screen.

Category	Details
Filter by Keywords	Helps you search, filter, and review conversations using specific keywords or phrases. Supports included/excluded keywords and speaker-based filtering, with results linked directly to the QA audit screen.
Conditions (AND / OR)	Select how the system evaluates multiple included keywords. - AND: All specified keywords must appear in the conversation. For example: Include `Billing AND frustrated` (both speakers) and Exclude `Resolved` (both speakers). - OR: At least one keyword must display. For example, include `Refund OR Money Back` (customer only) and Exclude `Processed` (agent only), and `Completed` (both speakers).
Conversation Including Keywords	Add keywords to include conversations in results. Select a speaker (Agent only, Customer only, or Both) and use + Add Keyword to include up to 8 keywords.
Conversation Excluding Keywords	Add keywords to exclude conversations from results. Select speaker scope (Agent only, Customer only, or Both) and use + Add Exclude Keyword to add up to 5 exclusions. All exclusions follow AND NOT logic.
Audit Navigation and Filtering	Enables quick filtering and navigation by keyword or QA question, with auto-scrolling of transcripts and relevance-based matching.
Apply Keywords Action	Applies the selected include or exclude keyword filters and updates results.

Filter by Custom Fields

This helps to refine interactions using business-specific metadata from Express File or Agent AI integrations. Custom fields are dynamically available based on the selected date range. The system excludes interactions that doesn’t include the selected field from results. To apply a custom field filter:

Select Custom Fields from the Filter By panel.
Select the Key from the dropdown. For example: Account ID.
Enter the matching Value.
(Optional) Select + Add Custom Field to include additional key-value pairs.
Select Apply or Save & Apply.
Conversations that don’t carry the selected custom field are excluded from results entirely.

Filtered Keywords Interactions Displays the count of interactions matching the selected keywords (for example, Filtered Interactions: 22).

Connector-Dependent Filters

The availability of certain filters depends on the Connector configuration. When Quality AI Express is disabled under Conversation Sources, the following filters are disabled:

By Experience: Avg. Waiting Time, AHT, and No. of Transfers
By Efficiency: CSAT and Avg. Waiting Time

When Quality AI Express is enabled, these filters are available when supported by the Connector configuration. The system automatically disables filters that the current Connector configuration doesn’t support, so you can’t apply them.

Manage Saved Filters

Use the Saved Filters UI to apply, store, and manage filter configurations on the Conversation Mining dashboard.

Action	Description
Apply Filters	Stores current settings as an Unsaved Filter.
Save & Apply	Saves and names filters for audit allocation. Saving is optional if audits aren’t used.
Save Filters	Saves the current filter configuration with a name and optionally sets it as the default on the Interactions dashboard.
View Saved Filters	Shows all saved and unsaved filters.
Filtered Interactions	Displays a real-time count of matching conversations.
Customization Options	Includes Copy, Mark as Default, Edit, and Delete filters.

If interactions aren’t used for audit allocation, you can apply the current filters without saving them as a reusable saved filter.

​Contact Direction

​Interactions Page

​Key Capabilities

​Conversation Records

​AI Agent → Human Agent Handoff

​Record Behavior

​Processed Tab

​Processing Status

​Access Process Tab Conversations

​Filter by Processing Status

​Conversation Grid Details

​AI Agent Conversation Behavior

​Pending Tab

​Access Pending Conversations

​Pending Tab Columns

​Filter By Processing Status Values

​Filter Conversations

​Add New Filter

​Filter Categories

​Filter by Efficiency

​Handled By

​Channels

​Audit Status

​Audit Allocations

​Duration Status

​Session ID

​Conversation ID

​Custom Conversation ID

​Handled AI Agent

​Experience Flow

​Human Agent

​Queues

​Agents

​Combined AI Agent and Human Agent Filters

​Advanced Performance Filters

​Filter by Experience

​Filter by Behavior

​Filter by Keywords

​Filter by Custom Fields

​Connector-Dependent Filters

​Manage Saved Filters

Contact Direction

Interactions Page

Key Capabilities

Conversation Records

AI Agent → Human Agent Handoff

Record Behavior

Processed Tab

Processing Status

Access Process Tab Conversations

Filter by Processing Status

Conversation Grid Details

AI Agent Conversation Behavior

Pending Tab

Access Pending Conversations

Pending Tab Columns

Filter By Processing Status Values

Filter Conversations

Add New Filter

Filter Categories

Filter by Efficiency

Handled By

Channels

Audit Status

Audit Allocations

Duration Status

Session ID

Conversation ID

Custom Conversation ID

Handled AI Agent

Experience Flow

Human Agent

Queues

Agents

Combined AI Agent and Human Agent Filters

Advanced Performance Filters

Filter by Experience

Filter by Behavior

Filter by Keywords

Filter by Custom Fields

Connector-Dependent Filters

Manage Saved Filters