Conversation Analysis
Conversation Analysis uses AI to generate insights from your agent’s conversations, helping you understand patterns, identify issues, and optimize performance. Analyze up to 250 conversations at once with intelligent filtering by date range and users.What is Conversation Analysis?
Conversation Analysis is an AI-powered feature in the Analytics dashboard that examines your conversations and generates actionable insights:AI-Powered Insights
Automatically identify patterns, trends, and issues using AI
Smart Filtering
Analyze specific time periods or users with flexible filters
250 Conversation Limit
Process up to 250 conversations per analysis for optimal performance
Secure Storage
Analysis results are stored securely for future reference
How to Run an Analysis
Configure Filters
Set your analysis criteria:
- Date Range: Select the time period to analyze
- User Filter: Choose “All Users” or specific user
Configure Custom Keywords (Optional)
Click Advanced Settings dropdown to specify custom keywords separated by commasExample:
inteligencia artificial, accesibilidad, eLearning, aprendizajeAdvanced Settings: Custom Keywords
Custom Keywords allow you to track specific terms in addition to automatically discovered keywords. The analyzer searches for these keywords in both user messages (INPUT) and agent responses (OUTPUT).
- Click the “Advanced Settings” dropdown
- Enter keywords separated by commas
- Example:
inteligencia artificial, accesibilidad, I&D, eLearning, aprendizaje
- Occurrence count: Total times the keyword appeared
- Relevance percentage: Proportion of conversations where keyword appears
Saved Analytics Reports
All conversation analyses are automatically saved for future reference.Accessing Saved Reports
The “Analytics Reports” section displays all your saved reports in a table format:| Column | Description |
|---|---|
| Report Name | Auto-generated timestamp (e.g., “Report 1/28/2026, 7:38:08 AM”) |
| Conversations | Number of conversations analyzed (e.g., 17, 20) |
| Date Range | Period analyzed with calendar icon |
| Load | Button to view the report results |
| Actions | Menu (⋮) for managing the report |
Managing Reports
Load Reports
Click “Load” to open any saved report and review historical insights
Delete Reports
Use the actions menu (⋮) to delete old reports and free up storage
Analysis Results
When you run a conversation analysis, the system generates a comprehensive Results report:Analysis Summary
Total Conversations
Total count with success/failure indicators
Processing Metrics
Average processing time, duration, and satisfaction rate
- Avg Messages: Message count per conversation
- Resolution Rate: Percentage of conversations resolved
- Avg Duration: Time per conversation
- Satisfaction: User satisfaction percentage
Sentiment Distribution
Visual breakdown showing:- Positive: Percentage of positive sentiment (green indicator)
- Neutral: Percentage of neutral sentiment (gray indicator)
- Negative: Percentage of negative sentiment (red indicator)
Top Input Keywords
Most common keywords in user messages, automatically identified by the AI: Example keywords:- equipo de Actúa
- Articulate 360
- IA
- equipo
- formación digital
Top Output Keywords
Most common keywords in agent responses: Example keywords:- variables
- eLearning
- inteligencia artificial
- personajes
- Storyline 360
Input/Output Custom Keywords
If you specified custom keywords in Advanced Settings, they appear here with detailed metrics.
- inteligencia artificial: 5 | 100%
- accesibilidad: 3 | 85%
- eLearning: 7 | 35%
Understanding Relevance Percentage
Relevance is a normalized value (0-100%) indicating in what proportion of total conversations a keyword appears. Formula: relevance = (count / totalConversations) × 100 Where:- count = Number of conversations containing the keyword
- totalConversations = Total conversations analyzed
- Keyword “servicios” appears in 25 conversations → Relevance = 25%
- Keyword “soporte” appears in 60 conversations → Relevance = 60%
- Fair Comparison: Compare keyword importance independently of total volume
- Relative Perspective: Understand how common a keyword is in your dataset
- Intuitive Visualization: Percentage format (0-100%) is easy to interpret
- Scalability: Works consistently with any number of conversations
Main Topics
High-level conversation themes identified by AI: Example topics:- company information
- technical support
- product training
- product information
- eLearning Content Creation
Top Intents
User intent classification showing what users want to accomplish: Example intents:- request information
- provide information
- request information about team
- request information about courses
Issues Detected
Issues include:- information unavailability (with severity level)
- Error patterns
- Failed tool executions
- Misunderstood queries
- Incomplete responses
Using Custom Keywords Effectively
Filtering for Better Analysis
Date Range Filtering
- Recent Activity
- Monthly Analysis
- Extended Period
Last 7-30 days - Best for current performance monitoring and quick health checks
User Filtering
| Filter Option | Use Case | Best For |
|---|---|---|
| All Users | Organization-wide analysis | General performance, broad trends |
| Specific User | Individual user tracking | Support cases, VIP customers |
| User Cohorts | Group behavior analysis | Beta testers, user segments |
Working with the 250 Conversation Limit
Strategies for Large Datasets
Time-Based Segmentation: Break into smaller time periods (e.g., analyze week by week), then compare insights across periods. User-Based Segmentation: Run separate analyses for different user groups, then compare results. Priority-Based Approach: Filter for the most important conversations (e.g., failed conversations, high-value users, specific issues).Use Cases
Quality Assurance
- Review AI-identified problematic conversations
- Spot response quality issues
- Validate agent behavior
- Check guardrail effectiveness
Performance Optimization
- Find bottlenecks and inefficiencies
- Optimize response strategies
- Identify successful patterns
- Reduce error rates
User Research
- Discover common needs and questions
- Map user journeys
- Identify pain points
- Track feature usage
Compliance & Auditing
- Audit conversation handling
- Verify policy adherence
- Generate compliance reports
- Track data handling
Next Steps
Analytics Overview
Return to Analytics overview and dashboard guide
Live Mode
Learn about real-time activity monitoring
Agent Settings
Configure your agent for optimal performance
API Reference
Explore the Analytics API
Frequently Asked Questions
How long does analysis take?
How long does analysis take?
Analysis typically completes in 30-60 seconds depending on the number of conversations and their complexity.
Can I run multiple analyses simultaneously?
Can I run multiple analyses simultaneously?
You should wait for one analysis to complete before starting another. However, you can view previous analysis results at any time.
Are analysis results saved?
Are analysis results saved?
Yes, all analysis results are automatically stored and can be accessed from the Analytics Reports section.
Why do I see different insights each time?
Why do I see different insights each time?
AI analysis provides estimates that can vary between runs as the AI interprets conversations from slightly different perspectives. Focus on consistent patterns across multiple analyses.
What happens if I have fewer than 250 conversations?
What happens if I have fewer than 250 conversations?
The analysis will process all available conversations that match your filters. The 250 limit is a maximum, not a minimum.
Can I export analysis results?
Can I export analysis results?
Yes, analysis results can be exported in various formats for sharing with team members or further processing.
Does analysis work in Live Mode?
Does analysis work in Live Mode?
No, conversation analysis requires Live Mode to be OFF to ensure you’re analyzing a stable dataset.
How do custom keywords work?
How do custom keywords work?
Custom keywords are specific terms you define in Advanced Settings. The analyzer tracks these keywords in both user messages and agent responses, showing occurrence count and relevance percentage.